Wren, an open source modular software synthesizer for Windows

Currently working on

2020-04-23: Added a random walk shape type to the LFO modules. The step size for the walk can be set with the warp controls, a low warp value giving a small step.

Note: for the next release must update the skin stuff for this.
2020-04-26: Made it possible to turn off the patch hints (the ones popping up for knobs and buttonss and such). This to "solve" a nasty issue with wine where the hint windows "hang" and that I do not know how to properly solve. It is a setting in the Settings Screen F3.
2020-04-27: Added some extra controls to the OscTrig module to make it more suitable for Pulsar Synthesis like effects. It now has phase modulation, shape modulation and warp control just like the main Osc module has them.
2020-04-28: Added an output type selector on the Osc and OscTrig modules, to make them more suitable for AM (amplitude modulation) type modulation (instead of RM, or ring modulation).
2020-04-28: Added a modulatable repetition count to the WavePlayer2 module. This to be able to play bursts of short samples. This is also meant to make Pulsar Synthesis easier to accomplish.
2020-04-28: Made two extra inputs on the WavePlayTrig2 module for programmaticly switching the retrig– and the loop–mode.
2020-04-30: Made a new sequencer type (Seq16Rep). This is an extension of the seq16 type where a repetition count can be set for each step (1 to 256 counts). Also a trigger pattern can be selected from n(one - no trigger pulse), s(ingle - one pulse per step), a(ll - one pulse per count) and g(ate - one long pulse over all counts). The repetition count and pattern are modulatable per individual step.

The Individual trigger outputs (trigs) as well as the main trigger output (trig) will output the selected trigger pattern, the tr. chain output will trigger once for every next step being taken, this can be used for chaining up another squencer such that it can stay in sync.

Pulse length for trigger pulses can be changed by changing the duty cycle of the the incoming trigger pulse.
2020-04-28: Added a new range for envelope times, the vfast range. it runs from 50µs to 10ms. This applies to all modules that work with changeable envelope timing ranges, those are: PulseDelay, EnvARRetrig, EnvAHD and EnvADSR. The granulators still use fast, I'm considering changing that to vfast to make them more usable. Anyway, audio rate envelopes are pretty nice for Pulsar like Synthesis.
2020-05-01: Added two new signal types HighLogic (orange, audio rate) and HighLogicCtrl (dark green, control rate). These will be used on inputs only (so there will not be wires in these colors). The difference with the regular logic signals (yellow and lime) is that the latter switch for levels > 0.0 and the new types switch for levels > 0.5. This makes it easier for some logic inputs to be controlled from unipolar signals. There had been a couple of existing modules already having the new behaviour, for those modules the colors have been changed to the new colors to make the input type clear.
2020-05-01: Added ranges for granular times too now, it supports the fast and vfast modes (as they existed for envelope times already). The bloody things need more work though as they still sound no good.
2020-05-05: Added relative LFO rate control (r.rate) for setting a global LFO rate multiplier. This is currently not being used anywhere, decided to do MIDI syncing in a more modular way. Oops, correction .. it actually does work as a global LFO rate multiplier, but it does not change the scaling of the delay lines.
2020-05-05: Made some changes to the Tapper module. An extra control mode was added (LFO fast 0.254) to allow for faster tempos with an LFO set to fast range. Some internal clipping was removed to allow for larger BPM ranges. An external tap input was added so this module can now be auto-tapped from an LFO (prably set to run in BPM mode for exact BPM matches) or from a ClockGen module's * 16 or * 96 output. The order of the control modes in the selector was changed, so some older patches may not be compatible without manually correcting the mode(s).
2020-05-06: Made a MIDI clock receiver module, click on the image to see the docs for it. It is to be used for MIDI clock synchronization. The sync output will give a short blip once every measure (24 PPQN is assumed and 4/4 mode).

Had to modify the overal MIDI reception structure for this, will have to test (N)RPN receivers after these changes.
2020-05-06: Made a RateConverter module, click on the image to see the docs. It is to be used for synchronization of LFOs and delays.
2020-05-07: Added alpha and hysteresis modulation to the CuClipStereo module, just like the mono version already had it.
2020-05-10: On Sequencer reset (Seq16 and Seq16Rep) the first step should become active immediately. Changed some things for it, seems better now.
2020-05-13: Added rate multiplier modulation to the MidiClockIn and Tapper modules. The RateConverter module already had this.
2020-05-13: Added a sync divisor to the MidiClockIn module.
2020-05-13: Added XFade modulation to the Mux module.
2020-05-13: Added an inhibit (inh) input on the EnvAR and EnvARRetrig modules, when it is active the envelope will not be started. This is a HighLogic input, it will be active for signal levels > 0.5.
2020-05-14: Added output type selection for the Noise and NoiseTrig modules (bipolar, positive only etc.). Some are redundant, like the pair 'positive only' and 'inverted positive only', which would end up being 'identical' in the long run.
2020-05-15: Changed the grain modules to use a mixing buffer for sample insertions, this smooths out the discontinuity between the oldest and newest sample in the sample buffer and so makes grain playing less glitchy. I've set the mix area to be 16 samples long which seems to be sufficient, it is a bit of a trade off, a longer fade area makes it substantially slower (limiting the maximum grain count to be used).

Did that same thing for the GrainDelay module, but that one still seems a bit glitchy …

Also removed a bug where the grain duration could become zero, which caused a division by zero.
2020-05-15: Changed clipping of the main program window on program load to allow for it to be sized to the screen size (without it being maximized). It used to be clipped 20 pixels short of that (unless it was maximized).
2020-05-15: Removed the slave output from the RateConverter module, it was only passing on it's input – so not too much use for it. (note to self: some patches of the last 10 days will need repair for this).
2020-05-16: Had fukt up the (trig) noise for the regular bipolar output range, fixed that.
2020-05-16: Added 'dust' as a new random signal type for the Noise, NoiseTrig, NoiseLFO and NoiseLFOTrig modules. It consists of randomly timed spikes, the distribution over time is set by the color controls. This is after an idea obtained from SuperCollider – via the Hetrick Dust module for VCV. I've changed the implementation a bit to use two independent noise values for spike–distribution and –magnitude.
2020-05-16: Added a ValueSync module, a value synchronizer. It acts a bit like a sample and hold but it will pass on the value captured when the pulse input got active only when a sync signal was seen after that. The Q ouput will go on for 1ms after a sync occurred, the captured value will be held untill a new pulse and sync sequence was seen.
2020-05-17: Added 'crackle' as two new random signal types for the Noise, NoiseTrig, NoiseLFO and NoiseLFOTrig modules. There are two versions Crkn for the normal and Crkb for the broken mode. The ideas comes from the HetrickCV Crackle VCV module (and originally from SuperCollider).
2020-05-18: Added an active output on the MidiClockIn module, intially or after receiveing a MIDI Start or Continue message it will be on. It will go off on receiving a MIDI Stop message. When active is offthe latest calculated BPM value will be set to zero and there will be no further sync pulses on the output. However the control and slave outputs will be frozen to where they were, so everything controlled by it will keep running as it was. When started again the sync output will go active briefly.
2020-05-18: Fixed a bug in the global LFO rate setting. All displays of LFOs would react correctly but the actual tate was never acted upon. It now works as intended, but at some later stage I might also want to alter delay times and envelope times.
2020-05-19: Added sequence length modulation to the Seq16 and Seq16Rep modules. The parameter value is being multiplied by the preset number of steps and can make the actual step count both larger and smaller, however the range will still be 1 to 16 steps.
2020-05-21: For the ModGateSeq module the gate modulation inputs were changed to HighLogic (dark green, switching at 0.5) mode.
2020-05-21: Added some functionality to the slew module. Slew can be modulated now and up and down slew have been separated to allow for a different change rate in one direction than the other.

in version 2020.4.13.0

2020-02-28: Changed file formats to all (except MIDI files, which are binary) use UTF-8.

Worst case, that is when you have used non ASCII characters in file names or module names, this may cause Wren to not start up anymore or to not be able to quit. Save your current Wren directories before upgrading to this.

I've tried to program a fix for this to make things backwards compatible, but can not oversee all possible situations. In case you see any trouble please report ‐ what will fix it is removing the wren.ini file, but I'd prefer to provide a solution without history loss.

I ran into trouble with the previous version when I wanted to save a patch with a greek name which was not handled correctly. The new version of Wren handled all migration issues for me – YMMV … let me know please – and I'll fix it.
2020-04-13: Made builtin MIDI, making the NetMidi program obsolete.

MIDI port selection is trough a new menu item setup/MIDI devices (the audio setup menu item was renamed to setup/Audio devices). MIDI Channel selection and (N)RPN preference setting still works trough the Settings Screen F3.
2020-01-18: Made steps modulation on the Quantizer module.
2020-01-18: Made step value modulation to the LUT module. This transforms it into something inbetween a constant step and a piecewise linear step processor (depending on the setting of the fade control). The modulation inputs are set to 1.0 when not connected so in that case the old generator only behaviour is maintained (modulation inputs are being multiplied by the slider set step values).

Also added an AM input, affecting all outputs but not acting on the chain input and also not active when the module is in Note mode.
2020-01-18: Changed 2010 to 2020 in places …
2020-01-22: A new module, a double pendulum simulation.

This resolves the [2017-08-12 : Double pendulum] request.

This module can either be free running or clocked, it will be clocked when something gets connected to the clk input. The m1 / m2 control sets the mass ratio between the masses at the ends of the pendulum arms. The sum of the masses is set fixed to 20 kg. The l1 / l2 control sets the length ratio between the lengths of the pendulum arms, the sum of the lengths is set fixed to 20m. The speed control and it's modulation input set the simulation step size, the simulation will not proceed when the speed is zero. For negative speeds the simulation will run in opposite direction (but gravity stays as it is). For large speeds the precission of the simulation will diminish and somewhat longer runs may result in non–chaotic behaviour. The reset input or button can bring the module back to it's starting state. With the G mod. input gravity can be changed, it can be made negative too, unconnected this input will have a value of 1.0 for standard downward gravity.

DPendulum animation example
2020-01-23: Changed mazes.4th to version 1.1 - just a color change really. Must not forget to use this one for the next release (it is in the web folder already but needs to be put into the zip-yet-to-be-created.
2020-01-23: There is an issue with the LUT module, xfade is not working correctly when the postion wraps back to the first one. Also outputs 2 and 4 seem to behave different than outputs 1 and 3? Hmm, nope, all ouputs act the same, so that bit is ok.

XFade is working correctly after all. As it can not look into the future it can not know that the position will wrap back in the next sample. So I've declared it to be a "feature" instead.
2020-01-23: I've changed the triangular wave forms to not use integration anymore but to use a BLAMP'ed trivial waveform instead. This behaves better with wave warping.

Changed the default bandlimiting to 1.004, close to 1 (which works best overall (it was set to 1.5 or so by default)).

Affected modules are Osc, MultPhaseOsc, MultiOsc, OscTrig and SquareTrig modules.
2020-01-27: A new module, a dual filter. Two 12 dB/Octave filters with three selectable configuarations and frequency spread control and pre/post distortion.

The two filters are connected in series, the first filter frequency is set with the frequency controls and the second filter follows it with a modulatable separation offset (which can be negative by using the separation modulation inputs).

The filter mode can be set to low pass (lp), high pass (hp) or band pass (bp). In low and high pass modes it will act as a 24 dB / oct. filter where the resonance peak can be split in two by the separation controls. In band pass mode there are two 12 dB / oct. filters where the bandwidth is controlled with the separation controls.

The lev control sets the input level and there is pre and post filter distortion available.

The Q control sets the filter Q for both filters, and it can be modulated positively and negatively trough the two modulation inputs.
2020-01-30: Added a divider on the clock out of the Seq16 module. I'm using this for chaining Seq16 modules at a different rate than it's "natural" rate (as set with the number of steps). This would save an extra divider module.
2020-01-31: Added a multiplier input to the PitchShift module, it is multiplied by the shift parameter to, for instance, allow for quick zero to set value jumps or for set value to minus set value jumps.
2020-01-31: Added "add" inputs to the Constant and MConst modules. They can provide a value to be added to the output value.
2020-02-13: A new module, a fast mode LFO. It is fast in that it is computed at audio rate instead of control rate, but otherwise it is the same as the LFO module. This module can be used to control delay line time modulatuion with less zippering than would be caused by using the standard LFO.
2020-03-05: A new module that does … nothing … It is handy at times for routing wires. and it adds one sample of delay. Just a bit cheaper than a mixer or something you could use for the same purpose. Besides, every program environment needs it's nop 😉
2020-03-19: A genome player module, translating RNA strands into control signals. This module scans text files for A, C, G and T (or U) characters (case insensitive) and builds a sequence from those with the letters mapped to the numbers 0, 1, 2 and 3 (such that A and T (U) are complemental, as are C and G). U is treated as if it were T.

Additionally W, S, M, K, R, Y, B, D, H, V and N are recognized, these are (according to their rules) randomly translated in one of A, C, G or T. W into A or T; S into C or G; M into A or C; K into G or T; R into A or G; Y into C or T; B into C, G or T; D into A, G or T; H into A, C or T; V into A, C or G; N into A, C, G or T.

All other characters (including Z, white space and numerals) will be skipped.

On playback the letters can be chained into groups of size 1 to 7 (3 would be the central dogma, codons encoding for amino acids (of which some may not exist for random letter sequences, but this module will assume existing sequences)). For a single letter there are four values [0.0, 0.25, 0.5, 0.75] or [0,0, 0,33, 0.67, 1.0] on the scaled output; for a group of two the values woud be [0.0, 0.125, 0.25, 0.375, 0.5. 0.625, 0.75, 0.875] (or their scaled equivalents).

On each trig the next group will be output, when the unit is in repeat mode it will jump back to the start when everything was played and when it is in once mode it will stop at the end. The res input will reset the sequence to the beginning. The changed output will fire when a value change on the output was seen. The leftmost outputs are complement outputs, for ease of implementation just 1.0 - output.

That sars2 thingie was a tad boring though to listen to … oh well …

in version 2020.1.18.0

2020-01-16: Embarrassing bug … there are anti aliasing errors in various oscillator modules when PWM or Warp is not set tp 50% (extra aliasing was added instead …). Also the PWM modulation does not work as intended on the PWM enabled modules. The affected modules are Osc, Square, OscTrig, SquareTrig, MultiPhaseOsc and MultiOsc.

Fixed all of this but the MultiOsc lost its PWM control in the process. It did hower get warp control instead which does PWM too except that the parameter names are different so some repatching will be needed on that.

Checked LFO PWM as well, which seems to be fine.

The Square, Saw and Tri waves sound a lot better now and PWM works as intended.
2020-01-09: Added solo buttons and a solo chain input and output to the MixS2to1 module. Making solo active on the left or right pair will block the audio chain inputs and and it will also block the other audio input pair. Also the solo chain output will then be active which is signalled by a yellow light. When the solo output is connected to a downstream MixS2to1 module that module will block it's local inputs and it will pass on only it's chain inputs. This will be signalled on that module by having an orange solo light.

Did the same thing for the MixS5to1 module.
2020-01-10: Added add, subtract an multiply modes for the A and R times on the EnvARRetrig modules. These control how the modulation inputs will act. In add mode, which is how things currently work and which will be the default setting, the modulation input value is added to the knob value to determince the actual time used. In subtract mode the modulation will be subtracted from the knob value and in multiply mode (which is how the modulation once a long time ago worked - and I wanted to have that again) the modulation input will be multiplied with the knob setting to determine the actual time.
2020-01-13: Added wave morphing to the OSC module. The morph input signal is being multiplied by four and then added to the wave selector control value. From this two waves are selected ( sine, tri), ( tri, saw), (saw, square) or (square, sine) and a mix from those two will be sent to the output, proportionally to the morph input value. It is the red input left of the wave selector.
2020-01-15: Added scaled clipping control to the RandSigs module. This to be able to make it spend more time at it's extreme values, which may result in more interesting volume patterns, for example (when used for volume control).
2020-01-16: Added pluck level modulation (AM) to Karplus-Strong based modules (KS and Pluck); Added AM input to HrastFilter (and made it a bit less high - pixel wise); Added AM to the HrastOsc. Added AM on SVF flter input; Moog filter, idem. Also AM now for Square, OscTrig, SquareTrig modules.
2020-01-09: Added a modulated gate sequencer, the modulation inputs can enable or disable a sequencer step. That is, a step manually set to one of the active modes can be turned off by setting the mdulation to zero or smaller. When the modulation is above zero a step set active by the buttons will be made active. Unconnected modulation inputs are set to one, so then the step will follow the button setting.

Steps that are turned off trough the modulation inputs will flash orange instead of yellow. They will not be skipped, but they will not pulse the output.
2020-01-15: Added a scaled clipper. The input values are first being level scaled according to the input type setting. Then they are being multiplied by the set level (level knob setting multiplied by the level modulation input) and clipped into a [-1.0,1.0] range. Finally the signal is being rescaled according to the selected output type.

The idea is that signals clipped this way will spend more time at their extreme values, i.e. a saw shaped input will be output as a trapezium shaped wave. But it can be used for all sorts of input values to make them spend more time at the extreme values.

in version 2020.1.6.0

2020-01-03: There are some issues with the TextWriter module (considered to be fixed now):

Changed the TextWriter module to allow it to only write into an existing folder. Also made it so that the text file will now be rewritten when the prefix settings changed.

When the folder would not exist (or maybe be located on a disconnected network drive) patch changes would get very slow while windows was looking to find the folder. The requirement for the folder to exist is assumed to solve this issue.

This will not work when windows still thinks the drive to be present, which happens after network issues. Sometimes it will come back, sometimes it will not in which case opening the drive in explorer will help – mostly.

Added an option to the Settings Screen F3 to allow the TextWriter module to work at all. By default the mechanism is disabled, so if you've used the TextWriter module before you'll now have to explicitly enable it.
2020-01-06: Delay times for the DelayMix module were not set properly, fixed.
2020-01-06: Added three extra delay ranges for the Delay, DelayMix, DelayStereo, Delay8, RndGranulator, Granulator, SimpleGranulator, GrainDelay, Looper and StereoLooper modules. Actually the ranges are the same, but there are 8192 steps instead off the 256 for the existing ranges. This allows for more precission in setting the delay time. All displayed time will show one or more extra digits too now, not only for the delay modules but for all modules displaying time.
2020-01-06: Upped the patch version to 9 to handle the changes made for the delay times properly. Older version patches should still be read properly. The displayed values though will have one or more extra digits, so things may look a bit different (like 110 ms in the medium range will now be displayed as 109.9 ms, but it still is the same actual delay time). This affects not only the delay based modules, but all modules capable of displaying time (like LFOs and envelopes, for instance). I've changed the display widths a bit for the longer representation to fit.

in version 2020.1.1.0

Sometimes it would happen that wire speed fixup was not processed correctly. The ouput module, for instance, would have blue input wires making things sound bad (as the signal would be control-rate then). Put a patch on this which seems to fix it, but it is a bit odd all … it depends on the order wires are drawn, a blue <- blue <- blue <- blue <- red chain should be transferred into an all red chain, but sometimes the leftmost two would stay blue.

As a temporary fix for current versions this can usually be solved by adding a module and removing it again (maybe twice), or by redrawing the affected wire connections. But the issue may reappear when the patch is loaded again.
There is a bug in the HardClip module, and maybe in other modules having the Lock option too. When Lock is set to off, on reloading the module the high level will be set to the inverse low level (which should only happen when Lock is set to Mirror Mode. In the Off mode the high an low levels should be independent.

Solved it by setting the default pairing mode to off instead of mirrored. All other modules with paired knobs already had the default mode set to off. Things will work okay when the pairing mode is turned on in the patch (on the module) itself.
When on the module selector the number of tab button rows gets larger than two, modules become mostly invisible and can no longer be selected. To change this I have made the program check for the number of rows and when it gets larger than two the width of the selector will be accomodated to reduce the number of rows to two.

This will not change the layout when the number of rows was one or two.
Fixed a bug in the "new patchname template" where the $N macro was not treated properly.
Changed the way module registration conflicts are handled. This used to cause the ini file to be corrupted, not so anymore (an issue encountered mostly during development of new modules).
Fixed the text color for the TEdit components on the Settings Screen F3, to make the text more readable. Default is white on grey now, the actual colors are user selectable (edit font color, under colors).
Made it possible to hide groups of controls from the toolbar. When made vissibe they will still be in a pre‐determined order. This can be set from the Settings Screen F3, but also there is a popup menu for it on the tool bar as well as there are items for it in the view menu.
Got a bit much going on in the OSC tab in the module selector, split it up into OSC1 and OSC2 tabs. Also rearanged the oscillators a bit better by category.
Added a freeze input to the Delay module. It will mix the fixed output and the input and that mix will be the actual input to be recorded into the delay line. When unconnected the module behaves as it ever did.
Added A, D, S and R lights on the ADSR module, likewise for A, H and D on the AHD one. And then added A, H and R lights on the two AR type envelopes as well.
Added an amplitude modulation (am) input to the GVerb module. When not connected it's value is 1, and it is being multiplied by the in level value to set the input attenuation.
Added a "1" output on the GateSeq module, it will be active when the sequencer is on it's first step.
Made a shuffle input on the seq16 module, it will randomly swap some slider values when activated.
Added an End Of Cycle (eoc) output to the EnvARRetrig module. Feeding it back into the trigger input will turn it into an LFO. It can also be used to make a sequence of triggered envelopes. The eoc output will only fire when the envelope curve was completed, so when the module is triggered too fast for completion to occur (it retriggers) this output will not go active.

Added the eoc output to the EnvAR module as well. This module will not retrigger so it will always complete it's cycle and so eoc will fire always.
Added an AM input to the filterbank modules.
Added a triggered wave player module, and a variation on it with variable start point and duration and some more control over the parameters. These are level triggered modules, the sample will start playing once the trigger input goes within the range set with the trigger level and the trigger range. This makes it possible to fire samples from a saw shaped LFO, each at its own time.

There is a loop mode setting allowing the wave to be played once or repeatedly (as long as the trigger condition is met). The trigger mode can be set to normal or to retriggerable ‐ in normal mode the wave will have to finish before it can restart, in retriggerable mode it can be restarted while playing by making the trigger condition false and then true again.

The playback speed can be controlled with the speed and the freq and/or fm controls, negative speeds (playing the wave backwards) are allowed too.
Added a wealth distribution simulation thingie. It has a pool of 1001 clients and on every internal update clock a transaction will be performed by two clients. Clients can transact only when they have more wealth than - kappa. When they transact the transaction will be some fraction of the wealth of the poorest of the two clients. This fraction ranges from 0% to 5%. When chi (the tax factor) is set to zero this will pretty quickly explode into an oligarchy where one client has all the wealth and the rest in maximum debt (as set by kappa). When chi is upped a fraction of the value of the transaction will be transferred from the richest to the poorest in the transaction and this will stabilize the system. The total internal wealth will be equal to 1001 times the set capital value (which is capital per capita), at least after a reset was issued (when tweaked wildly this may not be true).

On every input clock a random client is selected and it's wealth will be put on the output. Also a minimum and a maximum wealth is maintained, those values will (also when the clock goes active) be sent to the max and min outputs.

Now all of this really makes just a fancy random generator where small numbers will occur more often than large numbers (and where larger numbers will be spread out more in range as well), and as such it might be of musical use, especially when combined with the MinMaxRange module (see below) to make the output range more predictable. But also, it seems to make a nice rainy type of noise.

The internal update clock will either run at the sample rate or at the control rate, depending on the module being connected to a fast (red/yellow) or to slow (blue/green) modules only. The min and max values are recalculated once every 1001 internal clocks. For fast output clocking the min and max values may not always be accurate (as it could clock faster than the 1001 cycle period time). The numbers on the left show the current min and max wealth.

1001, a thousand and one nights she sat beside the sultan's bed. But also it is 7 x 11 x 13.
Made a MinMaxRange module. This is a scaler module for the case where the minimum and maximum values of a signal are known (as for instance in the Economy module above). The module will transfer a signal (sig) that moves within a range min to max into a range [0,1].

When the minimum value happens to be larger than the maximum value the module will still return valid output values, but it will work inversely (where the largest input value will map to zero instead of one).

Signals that are outside the [min,max] range will also be handled but will end up as an output signal outside the range [0,1] .

When the minimum and maximum values are almost equal the minimum value will be returned for whatever the signal value happens to be.
Made an equal division quantizer. An "octave" can be set which will be divided into the set number of divisions, all intervals thus created being equal in size. An "octave" of 2 is the usual octave, and 12 divisions will then create the usual 12EDO, 12TET or chromatic scale. The input range can be trimmed down with the range controls, these are common for all four channels.

The lights will indicate errors in the entered values (i.e. values that are not understood by Wren.
Made a replica of the Mutable Instruments Branches module, or the Bernoulli gate as it is also named. The Wren version has four gates, however they share a common mode- and p-parameter.

The module has four modes of operation:
- normal: on a rising input edge with probability p the topmost output will be turned on and otherwise it will be the bottom one. On a falling edge both outputs are being turned off.
- toggle: on a rising input edge with probability p the inactive output will be turned on and otherwise it will be the active one again. On a falling edge both outputs will be turned off.
- lnormal: or latched normal, as the normal mode, except the currently active output will stay active until the other one gets active.
- ltoggle: or latched toggle, as the toggle mode, except the currently active output will stay active until the other one gets active.
Made the output indicators bi-color, as the Mutable one has them. Dark means both outputs are off, yellow means the top output is active and orange indicates the bottom one being active.

The probability p is determined by the sum of the p knob and the p input. The resulting value will be clipped to a [0,1] interval.
Made a square wave rate multiplier. It Takes a square shaped input and then turns every edge in it into a pulse. It does not try to square up the signal, so the duty cycle information is lost. Modules can be cascaded to get 4, 8 etc. rate multiplications (but the pulses will not be spread out evenly then).

in version 2019.11.12.0

2019-11-12: Oops … made a bug in the WavePlayer which causes it to not always play, fixed it for version 2019.11.12.0.
2019-11-12: An undocumented new module for Gabe to check out.

in version 2019.11.11.0

Fixed a bug causing occasional memory allocaton errors in the eSpeak module.

Added a check to see if ESpeak had actually been installed by reading HKEY_LOCAL_MACHINE\SOFTWARE\Classes\TypeLib\{7192AA2F-F759-43E9-91E7-226371EF6B2F}\1.0\0\win32 from the registry, when that key is present and it's (default) value points to an existing file, it is assumed ESpeak was installed (at one time at least …). I'm not sure this will always work … this key is defined in the ESpeak source code for version 1.48.04, but it might break on later ESpeak versions.
Fixed a bug in the ProgDivider module which caused it to set out of range division values under some conditions. Also the divider could get out of sync when the counts were changed, that is it would get triggered on the oposite input edge, fixed this too.
Fixed the multiplex module such that it will no longer go out of range and also fixed the crossfading for it to work properly when less than the maximum amount of steps is set. The demultiplexer had the same issues, fixed those too.

Added a changed output to both the mux and demux modules which will generate a short trigger pulse whenever the selected step changes.
Removed a handful of range errors and divisions by zero and such. Did not run with range, overflow etc. checking enabled for a a good while ‐ I should have! (after all it's Pascal :-)
It seems to not be possible to have a single quote character in a patch name. Something to look into. Ok, this was a patch writer issue it did not do proper quote escapement (the patch reader already did it right). The same problem would arise on module titles with quotes in it. I've fixed this for both cases. User typed string values in edit controls were properly escaped already. Now there still may be an issue with multi‐line string values … k … the patch reader would not read line‐feed characters, which is fine for everything except for the value popup editor ‐ fixed this as well.
When oscillators are copied from a patch into another one in another instance of Wren with a different tuning being set, the new tuning is either not being applied or not visible in the user interface. Currently the only way to fix it seems to be to alter the tuning and then undo that. This will probably apply to all tuning sensitive modules, filters for instance. Indeed, the tuning of the imported modules is applied to the patch instead of the tuning of the patch being applied to the imported modules. This will probably go wrong with importing saved module clusters from file as well. Fixed it, only loading a completely new patch will affect the current tuning parameters, and none of the pasting or module read options will do so.
Fixed a "bug" on the Looper and StereoLooper module where it would not output any sound until the speed knob changed position. Also changed the speed input for the WavePlayer module to have a default value of 1 (to make the wave being played without any furter connections, when you'd want a "standing wave" connect the speed input to a constant zero value).

Later I've added a selector for the speed mode, by default it is on which makes the wave to be played at regular speed. When turned off it makes the wave stand still. When the speed input is connected to something however this will override the speed mode setting. So you can now have it function in both the old and the new way.

When the wave stands still it can still be played by varying the pos input, for instance by connecting a saw shaped LFO to it.
Fixed a bug for the Constant, SeqValues, Lut and Knobs4 modules where, when the mode was set to "note", the value knobs and sliders affected by this mode would loose their value on module copy or patch reload. These were all cases where the step count of the associated control would change with the selected mode.
Fixed a bug for the SeqValues module where wrongly colored dots could be set on the P, S and E controls.
Changed the Vocoder FFT overlap from 75% to 50%, with the Hann window used 75% created some nasty squeals at 1/3 of the sample rate (which of course I didn't hear in my old ear ;-) ‐ but the spectral analyzer …

And while at it, I made the FFT size user selectable. A larger size first will sound better afte which it tends to smear things, but also a larger size will give more processing delay.
Changed the WavePlayer module to always play at the same rate, indepent of the current sample rate. It still only accepts 44k1 stereo or mono wav files.
Made the VU meter update rate independent of the lights update rate, it is fixed now at a 23.8Hz rate.
Added two macro's to the TextWriter module. %patchname% will be substituted by the name of the current patch and %prefix% will be replaced by the TextWriter prefix value, which can be set trough the Settings Screen F3 ( when prepend prefix is checked it will be prepended always, when it is not checked it will be prepended only when the %prefix% macro is used in the TextWriter module ‐ the prefix value itself can be left blank too when no prefix is wanted).
Added an active input to the eSpeaker, when active speech will proceed, when not it will be frozen in-place. With an SR flip-flop triggered from an LFO and reset from, for instance, the word trigger out of the module, speech onset can be synced to the LFO. It will still be async while it runs, and this will cause clicks due to the abrupt starting and stopping, but at least it's synced a bit.
Added inverted outputs to the RangeConverter module, where the regular ones go up the inverted ones go down (over the same range), and vice versa.
Added changed outputs (in the column marked "C") to the seq16 module. These will go active briefly when the associated sequencer output changes.

And also added clear inputs for "one", "some" and "all", similar to the randomizer. The some inputs will clear or randomize four steps that are separated by four steps. The starting step is chosen randomly from the first four inputs.

Added three extra chain inputs, to make it possible to chain all four sequencer lines. Some patches may need modification (connect the original one chain input to the three new ones) for it to function the same way as it did before. This is the columne of inputs marked with "c"
Added an "am" (amplitude modulation) input to the basic Osc module.
Added an "am" (amplitude modulation) input to the Looper and StereoLooper modules.
Added shape warping to the basic LFO module. This acts like PWM for a square output, a triangle can be morphed from saw to ramp, etc. This works for all the wave forms, but the random waves will not be much different on warping.
Added a range selector to the ARRetrig envelope module. This allows for way longer times to be set for it. For patch compatibility it will come up with the default fast mode (it always had).
Added module feedback to the edit history window, a selected history item will now highlight the associated module in the patch editor.
Added a brownian noise mode to various random generators.
Added a Filter mode to the PulseDelay module, this will cause it to only be able to re‐fire after the input was inactive for a couple of sample times ( depending on the control rate divisor). The ouput is quaranteed to stay low for a couple of sample times after the stretch period ended. This makes it possible to set a maximum rate for the output to fire. For instance, when the input rate is 1Hz and the filter time is set to 2.5s the ouput will fire no more often than once every 2.5s, so in this case it will fire once every three seconds. This differs from the stretch mode where with the same settings the ouput would never go low.
Added a PADSynth like module, it use IFFT to generate sound from a spectrum which is created by a set of faders and some extra controls for spread, peak bandwidth and bandwidth scaling with frequency. Some of the controls are clicky as the IFFT has to be recalculated (and I was too lazy to make fades for that). Also this does not fully implement the PADSynth, it just uses it's basic ideas as laid out by Nasca Octavian Paul ‐ anyways, it can make some nice sounds.

The faders set the spectrum, the bw (bandwidth) controls set the width of the spectral peaks.

The spread parameter manipulate how the spectral peaks will be spread.

The module works by generating an in‐memory waveform from the spectrum. This waveform can be played at different rates and from different positions by using the position speed and frequency controls. At some stage I should modify this to have multiple wave forms to be generated with cross fading implemented over that, too lazy for that now.

The module has a stereo output, a diffrent tap point is used that for that on the calculated wave.

The IFFT size to be used can be set, small values will act a bit weird, anything up from 8192 seems okay. Values from 64 to 262144 can be used.
The 300th module :-)

Added a new granulator type module, working better than the three existing ones. It has a delay line able to hold 5 minutes of audio and a maximum of 128 grains can be set (provided that the CPU can cope with that).

It has grain controls for pre-delay, spread (or grain distance), pan, duration, frequency, speed and amplitude. All these controls can be modulated. Additionally the amount of grains can be set from 1 to 128, and this can be modulated too.

Several grain envelope types can be set to change the sound character: bell, pulse, exp, reverse exp, tri and box, for all of these there is a noisy variation as well.
Added a smearing module … why? because I could. It smears the phase of the incoming signal with a controllable random amount and it adds some delay, FFT based.
Added a MinMaxMix module after an idea from an idea from Ricko at the electro-music.com forum. When both input signals are positive the ouput gives the largest one of them. When both are negative the output is the smallest one. When the signs are opposite the sum of both inputs will be sent to the output.
Added a PhaseSplitter module, it will split the incoming signal into two signals that are 90 degrees out of phase. This can be used as a building block for various audio effects. for instance the Bode frequency shifter internally makes use of the same splitter code.
Added a Detector module, it will split the incoming signal into two signals that are 90 degrees out of phase, thus "complixifying" it. It will then use the complex signal to derive instantaneous amplitus\de, phase and frequency. This works properly only for sine shaped signals, but also "sort of" (as in "nicely") for more complex signals. I've used it to analize a voice signal to control an oscillator to implement pitch shifts on that voice.
Added an AbsAdder module, it adds up positive input values only, all negative inputs are clipped to zero. This is a sort of a "Micky Mouse" logic "OR" function, the ouput will never go negative.

Added a DTMF tone generator. It's input value range 0 .. 1 is scaled into a range 0 .. 15 which then is used to select a symbol from [ 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, #, *]. The symbol then is converted into two tones according to the following table:

	1209 Hz	1336 Hz	1477 Hz	1633 Hz
697 Hz	1	2	3	A
770 Hz	4	5	6	B
852 Hz	7	8	9	C
941 Hz	*	0	#	D

Added two Xenakis like number sieve modules, one with a blue (addressing) input control and a clocked version.

Wren's notation for basic modules is:
- [n,m] : a sieve module with modulus n and residue m [3,1] -> {1,4,7...}.
- {numbers|n} : a litteral number sequence, f.i. {1,5,13,14,21|24} n is the modulus, 24 in this example.
- fibonacci : the fibonacci sequence {0,1,1,2,3,5,8,13...}.
- prime : the prime numbers {2,3,5,7,11,13...}.
- hanoi : the hanoi disc sequence {0,1,4,5,8...}.
- hanoipegs1 : hanoi peg sequence 1 - non alternating {0,1}, {0,1,3,5} , {0,1,3,5,6,6,7,8}.
- hanoipegs2 : hanoi peg sequence 2 - alternating {0,2}, {0,1,3,5}, {0,2,3,4,6,6,8,10}.
- hofstadterq : the Hofstadter Q sequence, chaotic.
- metafib(n) : genralization of hanoi, n being the order, a whole number.
- random(f) : a sequence of random numbers, f selects the chance for any number to be present in the sequence. 0 <= f <= 1, f.i random(0.33).
These can be combined using the following operators:
- + : the union of two sieves: [3,0]+[4,0] -> {0,3,4,6,8,9,12...}, the numbers combined (logical or function).
- * : the intersection of two sieves: [3,0]*[4.0] -> {0,12,24...}, the numbers present in both (logical and function).
- ^ : the symmetric difference of two sieves: [3,0]^[4,0] -> {3,4,6,8,9,15...}, the numbers in one but not in the other (logical xor function).
- ! : the complement of a sieve: ![3,0] -> {1,2,4,5,7...}, the numbers not in the sequence (logical not function).
Parenthesis can be used to enforce an evaluation order, f.i. as in (([3,0]+4,0)* fibonacci)+prime to first evaluate the left +, then the * and the right + last.

A simplified formula will be calculated when the formula changes, or when the external period changes. The simplification does not use prime decomposition, it will compute a sequence and from that it will make a sequence of modules joined through union.

The evaluate input will force reevaluation of the formula, this can be useful when a random module is present to obtain a new random sieve.

Otherwise this module functions like the Values module, except that the start value will always be 0 (zero) and the end value is set by the external period. This means the input value changing over the standard range as set up with the input range selector will address all sieve values.

The difference (top) output will output differences between sieve sequence values, in case the last sequence value was selected it will be the value of the last minus the first element (this cyclic behaviour was not present in the original concept).

Negative numbers can not be used in sieves and the random sieve's probability control is the only floating poin number being used. The ouptuts will still use float values, as they are scaled to standard output ranges ([0,1, [1,0], [-1,0], [0,-1], [-1,1] or [1,-1].

The simplified formula is displayed above the user supplied formula, or an error will be displayed when the user supplied formula could not be parsed.

The currently selected sequence and diffrence values are displayed in white, these are the unscaled values.

The changed output will go active when the ouput value was changed, the difference output may not have changed then.

The regualar (bottom) output will be scaled such that the external period results in an internal output of 1, this can be changed then with the output type selector.

The difference (top) output will be scaled such that the largest difference value present will result in an output value of 1, again this can be rescaled using the output type selector.

For more information on the subject try googling for Xenakis sieve, or Xenakis sieve algorithm. It may find a 221 page paper by Dimitrios Exarchos (Iannis Xenakis and Sieve Theory), I've found that to be a very useful text.

in version 2019.6.9.0

[obsolete] Fixed a bug with legacy audio, it would get stuck in a start / stop loop, broke that loop ‐ audio still seems to suck though in legacy mode. I've been considering removing legacy mode completely.

As of 2019-04-10 legacy audio mode was completely removed. Great ‐ this made patch recompilation work faster (legacy audio was started and stopped for recompiles (adding or removing modules or cables, or loading a new patch) but PortAudio is kept running on recompilation).
Fixed a bug for the ScaleQuantizer and the ProbSequencer modules where the initial scale after patch start was not shown correctly when scale modulation was used. It would not show the initially selected scale in the notes switches.
Fixed some of the most obviously wrong help items, had forgotten to update the text for some of the more recent modules.
Changed the patch format to include the tuning parameters that were set. The patch version did not change for this as the modification is both upwards and downwards compatible. Old versions of Wren will skip tuning parameters when present and new versions of Wren will not change the tuning when the tuning parameters are not present.

Tuning parameters will only be read when a new patch is loaded from disk, they will never be used on parameter imports and such. They will be read when a a patch template is loaded, as this is the same as loading a new patch from disk.

An option was added to the Settings Screen F3 to never import tuning parameters from a patch. When this is active tuning parameters are still read from the ini file when Wren is started, and tey will then be the same as when Wren was last closed. This option is on by default to be compatible with previous versions.

Also fixed the hint popup for knobs to show the right frequency after tuning changes were made. The note names are still incorrect, but I do not have ideas about how to properly name notes in general, for instance in a tuning of say 11.21 notes per "octave" sized 2.13, say …

I've brought the tuning controls to the main toolbar, they were in the Settings Screen F3 before.

Then had to change the patch version from 7 to 8 anyway as I did add a couple of new LFO ranges and wanted older patches to be read with correct range values and the variations to stay intact as well, see the next paragraph for more on this.
Added three new LFO ranges, being fast2, medium2 and slow2. These have ranges of values mapping nicer to whole time intervals (for instance a rate of 1 Hz can be set now) and also the mapping to standard oscillator frequencies is better (for instance an LFO rate of 55 Hz can be set now now, although this will only make sense in a standard 12 tone octave based tuning with an A of 440 Hz, as LFOs are not rescaled with the current tuning). Had to change the patch version number for this to be able to read in older patches correctly.

The new patch version is 8. There are 32 steps now for a change of a factor of two, for the existing fast, medium and slow ranges this is 24 steps. Which also means that the total span of the new ranges is a bit smaller, but of course the old ranges can still be used for the more extreme settings.
Changed the A and R time modulation behaviour for the EnvARRetrig envelope module, the modulation is now added instead of multiplied. Added A and R modulation for the EnvAR module as well, working the same as for EnvARRetrig. Also made the modulations to work in real time ‐ before they were sampled only when the envelope got triggerd. This will break old patches when they use time modulation for EnvARRetrig modules.
Added some new fuctions to the Monadic and Dyadic modules. To Monadic was added 1 - abs( x) and to Dyadic x ^ ( 1 / y). Also changed the behaviour for the case where the function can not be calculated (like division by zero, for instance). In the new situation the output value will not change from it's previous value in case of an impossebility (output used to be zero for the Monadic module and X for the Dyadic module in such cases). This may break some old patches that were based on "undefined behaviour". Also the correct function may not be loaded for old patches.
Added 32 row sum outputs to the LifeSeq module, just so the thing can control more and better. This was inspiered by PHOBoS, but he is using an 8 by 8 grid where each cell is a different note. This seemed a bit much for a 32 x 32 grid, so I modified the idea a bit. Also made a new 12 by 12 SmallLifeSeq, more symmetrical, with both a controllable horizontal and vertical cursor. Then also added yet another game of life module the VitaPHoBium. This one has a 16 x 8 grid where input values can be clocked in to turn cells on. This one also has X Y cursor control as well as stepped cursor control, and when the cursor lands on a life cell a note value of sorts will be output. Later added a "Skip dead" mode which will try to find the next live cell on a next clock. skipping the dead ones. When X and Y are connected but Step is not the cursor will be updated asynchroneously, and when Step is connected as well the eX and Y inputs will be sampled on an upgoing Step edge. When only Step is connected the cursor will move in a stepped way, either forwards or backwards.
Made an unclick module, this is just an averaging filter with a fixed alpha setting ‐ it is a bit cheaper than the average filter, and it will filter out the most annoying clicks without giving a lot of lag.
Added 'raw' outputs to the Seq16 module, they are raw in the sense that they are not cross faded. Marked the existing outputs with an X and the new ones with an R, for cross-faded and raw respectively. Did the same for the LUT module.
Added a changed output on the ScaleQuantizer module, it goes active briefly when the current output value is not equal to the previous one. Likwise for the poetry module.
Added inverted outputs in addition to the regular ones for the RandSigs module. The inverted outputs are using the same output type as selected for the regular outputs.
Changed the way scale modulation works on the ScaleQuantizer module, to make it look a bit less confusing maybe … For an unconnected or an invalid scale selection input the pre‐selected scale will be used. When the scale modulation is connected and it has a valid value the pre‐selected scale will not be used but the modulation in itself will determine the scale to be used. Before this the pre‐selected scale would be added as an offset to the scale modulation input. So there is no longer a need to set the pre‐selection to the first value in order to get predictable scale modulation. Also it is possible now to set the preset to a fallback value to be used in case the modulation goes out of range. In the old situation invalid modulation input could cause the input value to be passed unchanged to the output, this is no longer possible, so this change may break some older patches where this "flaw" was used as an effect.
Added effect modulation on the AmMod module. This makes it possible to programmatically fade between amplitude and ring modulation.
Made a Dif module, a differentiator, with a fixed gain of 1. There already was an IntDif module as well, but the integrator bit did not seem too useful to me … so made it a bit cheaper. The old IntDif module is still available as well.
Added a ChangeDetector module. It detects changes on it's input and it will send out a trigger pulse when a change happens. Also the value of the change will put onto an output, it can be selected whether the value of the change is taken as is or has it's absulote value being taken. This sort of is a differentiator with a built-in sample and hold function.
Made a voltage controlled pattern sequencer, it was modelled after an idea by fonik but made it a 16 step one and addded reverse and inverse modes as well as step length modulation and random control over the slider positions. The Gain and Offset controls work with bits, so they can be used to set the step where the effect begins and ends. Hmm … not quite … but will leave it as is for now.
Made an AutoQuantizer module, it samples input values clocked in on it's learn input. It will remember the note value (but not the octave value) of the last N ones, where N ranges from 1 to 64. Signals coming in on the (reqular) in input will be split into an octave and a note value and the nearest note value present in the history will then be output in the octave of the incoming value. This can sort of improvise with notes it had seen before by putting some sort of a random value into the in input. When a less random input is being used like a saw or a triangle it would output arpeggios. When the new output differs from the one before a trigger pulse will be generated on the changed output, this can be used to trigger an envelope, for instance.
Made a NoteSplit and a NoteMerge module. The splitter splits an incoming note value into it's ocatave and note parts. The merger does the reverse thing, it combines a note and an octave value into a note. The octave size is obtained from the global NotesPerOctave setting. The merger is really just an adder, but still, being the inverse of the splitter I'll just keep it.
Made a Queue module. It remembers the latest N clocked in values (with N going from 1 to 64). With the select controls one of the remembered values will be selected, optionally providing cross fading between old and new values.
YASM ... this speaker uses windows SAPI voices. Right, can only get it to work in 8 bit mode and it is very slow to translate speech to text. I've made a loading indicator which lights up while the conversion takes place in the background. This way it at least is possible to go on patching while the speech synth compiles the text. Funny though how it can reasonably quickly perform a voice change, text changes take much longer. Anyways, it is more like speech than the other speech labeled modules.

Windows is a bit limited in the number of voices it supports out of the box, to fix that eSpeak can be used. On installation you can select languages and speakers to be installed, some modifiers are supported too.

Ok, got the sluggishness out a bit, turned out it was forever appending and partially overwriting to the same audio queue instead of renewing it. So had to destroy and recreate it instead of reuse it. Still, longish texts at high rendering rates will take a while to render, like a couple of seconds for the text from YASM to here.
AYASM ... this speaker uses the eSpeak voices and the espeak-sapi.dll. This one is a bit slow to compile speech too (albeit a tad faster than the SAPI one maybe), for short text it will do. It more or less provides the same functionality as the older song module, except it has a better speech quality. The module will signal speech events like phonemes, words, sentences and silence. Also markers can be placed in the text to generate user events. The user events can be made to generate an output pulse, or they can put some value on an output. The number of voices is fairly large, and there are modifiers for male, female, whisper and croak. Also … it can speak Dutch with a propperly rrrroling godverdomme :-)

Too bad tho that the eSpeak provided voices seem somewhat arbitrarily tuned. Anyways put some tuning controls, inclusing a cents control, on the module to cope with that.

ESpeak should be installed for this module to work, and the espeak-sapi.dll should be somewhere in the path (it can be put into the wren executable filder, for instance), the standard windows DLL search policy applies.

I've added the eSpeak DLL to the wren package, but forgot to include it into the source zip I think.
Added cents values to be displayed in the popups for FM modulation on audio rate oscillators as the concept of semitones is a bit odd for some scalings.
Changed the way the "Freeze" indictor works on the Reverb module, it no longer shows the value of the Freeze input but the actual freeze status (which is the xor of the input and the mode selection button).
Made the patch compiler a bit more robust. It would on bad module or cable definitions crap out and not load the patch at all. It will now load the patch skipping the stuff not being understood. Any skips will be logged.
Removed the splash screen animations, it was uggly in more than one aspect …

in version 2019.3.5.0

Fixed a bug causing unnescessary crackling. Did this by totally removing all automation stuff ‐ it was not working anyway.
Fixed a bug with portaudio initialization where it was not possible to have an output only system. It is now possible to have an input-only system as well … how usefule is that ;-) But there need to at least be inputs or outputs for Wren to be able to do audio I/O.
Fixed a bug with reading wren.formants where the file could not be read on some windows systems with a code page differing from 1252.
Various small changes on several modules, and a couple of new ones. I didnt keep track of this process very well though, so … surprise.
Oh, the ones I do know are: Added a couple of new modules, mostly Chladni related, a generator a filter bank and an oscillator bank. Oh and the Van der Pol stuff … more info later. Also added some inputs here and there … like the "1" input on the Sequencer16 and LUT modules, these randomize one of the 16 steps (a randomly selected one) instead of all at once (which is now the "all" input). Also added an interpolating slow to fast mode conversion module, helping to get less crackle on delay line modulations when the control rate division is set high. Also made it possible to have "octaves" of a factor different from 2, this is in the Settings Screen F3.
Trying to port stuff over to free pascal … but that will be a while. First splitting off the engine, that would be Wrengine. It compiles and it succeeds on some unit tests (like building a simple patch and executing that). I've found DLLs for Windows, Linux and OSX too. Had it compile under all three OSes as well, but didnt do any 'real time' audio testing as of yet. Also the user interface will need to be completely redone. There may be some windows only releases still before this wil take off. Anyway, for the curious, this is available for download (in the downloads section at the top of this page), just note that it is not a synth yet but an experiment. now at wrengine.2018.5.19.
A little update on the above, it can now read Wren patches version 2 to 7. Not all modules were implemented for it yet though. It does not have the Swansong (there is an issue with the forth interpreter) and it does not have the Van der Pol modules (which were recently added to Wren). Now at wrengine.2018.5.22
And a lil bug fix for the variation values. Now at wrengine.2018.5.23.
Added variations, morphing and automation stuff, not finished (but consuming CPU time). Also added support for reading Wren patches and added some profiling code to check the performance of the synth engine ‐ not too bad sofar, better than the original Wren it seems:
```
    Running synth for 1000000 cycles
    Duration was 9099 ms : time / sample : 9.1 us, sample rate : 110 ks/s
          
```
Which was for a patch doing about 60% for Wren at 48 ks/s (or about 80 ks/s at 100% CPU, but no actual audio i/o was performed for Wrengine where it was for Wren), and optimization mildly set to o1 and many checks on. For o4 optimization, i/o checking, range checking, assertions and debug code off it was like:
```
    Running synth for 1000000 cycles
    Duration was 5672 ms : time / sample : 5.7 us, sample rate : 176 ks/s
          
```
For all cases the control rate decimation was set to 8.

Also some small bug fixes were made, should check all the p_ stuff to not use FInputs but FParams instead … now at wrengine.2018.5.25.

in version 2018.5.18.0

Added a matrix multiplier module. It calculates M_abcd * in → out (or a * in1 + b * in2 → out1 and c * in1 + d * in2 → out2).
Made a new modal filter able to load presets from a text file. The frequencies read will be fixed, but an overall tuning can be applied either exponentially or linearly. Individual Q and amplitude settings are present. Up to eleven formants can be specified and the overall Q factor can be set as well. The text file read is wren.modal, and it needs to be present in the Wren executable directory for this module to work.
Made a new MultiGate module providing four dual input gates all performing the same selectable function. The input and output signal types can be set separately.
Made a 3 x 32 / 1 x 96 sequencer. Notes can be set per step as well as an octave and a duration.

Interpretation of the duration depends on the mode set (either free or sync). In sync mode step durations are fixed to 16 clock periods and the gate time will be equal to the duration, while in free mode the step duration and the gate time are both set to the selected duration.

Trigger mode can be set to Trig, Gate or Repeat. Trig outputting a clock for 1/16 of the input clock period (which is 16 clocks long), Gate for the set duration and Repeat does repeating 1/16 clock pulses during the set duration.

The three button controlled modes can also be modulated.

Split the clock input into three separate ones so the three play pointers can move at different speeds. This will also work in 1 x 96 mode, where the first output will be the sequence proper and the other two will be canon like variations on it.

The dir input when active will reverse the playing direction, the inv input will play the sequence upside down when active. These are global controls working on all three play pointers the same way (just like act and res do).

The / 16 outputs output the incoming clock divided by 16.
A van der Pol oscillator module.
A van der Pol LFO module.
Added a drive input to the Attractor and AttractorLFO modules ‐ just like the van der Pol thingies have that.
Added Sin and Cos outputs to the rotator module. These can be used to control delay lines to get a leslie type effect. Also added a double sin mode ^* where the Cos output will actually be -Sin and where the rotation being berformed works a little different.

^[*] Or maybe it's zero sin with the minus involved.
Made a stereo version of the Delay module, DelayStereo. This is just two delays sharing the same delay length parameters.
Made the phase of the RandSig module to have a random value after reset. Also made reset to cause an immediate 'wrap' action (starting a new random cycle). Reset used to only set the current phase value to zero.
Retuned all of the fixed filter banks a bit in such a way that they will use 55, 110, 220, 440 Hz etc.
Added an option to the Pad mdule to have a user selectable background. Also added three new outputs refelcting the color of the image pixel the cursor dot is on.
Added four extra multiplication units to the MultipleMult module, giving it eight multipliers in total.
Some minor bug fixes.
I've moved from Delphi XE5 to XE10. The latter seems to be slowish .. but hey it's the fuuture, one can't stop that.
I'm totally rewriting the automation recorder. It wil use a flat model now with a track for each control, and no extra module level anymore. The user can select between recording all knobs or just a set of exported ones. Each track can have it's own graphical editor. It will also be possible to add or delete events trough that editor. Currently it is not working at all.

in version 2017.11.19.0

Fixed some issues with the history view and the alternate module titles. Like when a module was deleted it's history could still seemingly be undone, resulting in an error, of course. Deleted modules will be removed from the patch history now. When a new module gets inserted into the patch all modules will be renamed, this situation is handled correctly now by renaming the history items as well. When alternate module titles were active a newly inserted module would still show the regular one. Fixed that one as well.
Had over-optimized patch compilation a bit, resulting in patches to not always play correctly after load. A workaround is to insert a new module and then delete it again. Fixed now.
The Gate module was not functioning properly for a bit count of one it used two instead causing the negated functions nand, nor and xnor to not work and causing the regular functions and, or and xor to behave a bit odd. Fixed now. The one and none function modes were not affected by this.
Introducing Hold mode for the (non‐retriggrable) AR envelope module introduced an error making the envelope hang in the A phase, fixed that. Same thing for the retriggerable version.
Module initialization was changed a lot in the previous release, causing the filter bank modules to generate exceptions at initiailization time (they looked at inputs at a time inputs did not exist yet) ‐ fixed that.
Fixed an error in the all pass filter, it's actually working now.
Added some functionality to the randomization limiting. For all randomization it implements a scaling function now. For mutatation, mating and morphing it implements clipping. For clipping, when the maximum value is set lower than the minimum value (a red circle segment around the knob) clipping will be performed as exclusion, as in that the range marked as red is never being used. Likewise for randomization with reversed limits the mapping will be will be 'un-scaled', i.e. mapped to values outside the selected range. So there is no inversion anymore but exclusion.
Added a new morphing mechanism. There are four morphs available, and each knob can be assigned to one or more morphs (and it can have a randomization range on top of that too). The active mprph can be selected from the top toolbar. Moving the knob present there will move all the morphs in that group between their set ranges. I've also added a morph control module to be able to automate morphing. This mechanism is separate from the previously existing variation ‐ or auto ‐ morphing. Knobs having a range set upon them which is not the currently selected range will show a grey circle around them, a selected range will be either blue or red. The MorphController needs to be enabled in the toolbar with the (M)anual / (A)utomated switch, morphs are either manual or automated. The MorphControl module provides a base setting for the morph knob and some modulation to be applied. I've also added a switch between (A)utomated and (M)anual for the auto morph system (morphing between variations). When a morph is automated it can no longer be manually controlled, as indicated by a red dot in the controlling knob (locked indicator) ‐ as can be seen in the image above for the variation morphing knob above. For morphing there is no exclusion, it will use reversed scaling instead, i.e. morph selector going up results in knob going down.
Added PDOsc module, a phase distortion oscillator. It has a fixed sine shaped wave to which overtones can be added by changing the ratio parameter. It turns out that phase distortion done 'Casio style' is very similar to VOSIM. Anyway, this seems to be a nice oscillator for bass tones.
Added a multi phase oscillator. This is a replica of the main oscillator with three extra phase shifted outputs added. Phase shift can be set to 90 or 120 degrees ‐ when set to 120 the lowest output will be a 180 degrees shifted wave. All the wave forms of the main oscillator are present, and all outputs are anti‐aliased.
Added a multi phase LFO. This is a replica of the main LFO module with three extra phase shifted outputs added. Phase shift can be set to 90 or 120 degrees ‐ when set to 120 the righmost output will be a 180 degrees shifted wave. All the wave forms of the main LFO are present, but phase shifts on a random wave make not much sense to use (as they will be uncorrelated).
Added a MUSE like module, implemented from the schematics publised by PHOBoS on electro-music.com. I've left out the audio generation, the module outputs a control signal instead. I've added some extra features to it: two external inputs going to to the switches, two extra divisions, by five and by seven, added two carry outputs where the overflow from the shift register ends up. Also added an inverted output signal, and the shift register's direction can be reversed.

The free / clocked selector controls how the output follows changes in the interval selection, they can be synced to the clock or immediate (in the schematic linked to above the changes are immediate, but when using switch selection modulation it is handy to have the changes occur synced). In synced mode the res input will be synchronized to the trig input as well. The parity generator can be switched between odd and even parity. All actions related to the shift register (parity generation, direction reversal and the W … Z switches) are inherently synced to the trig input. Using an ext input and a carry output two or more MUSE modules can be made to interact with each other.
Added a 32 bit cascadable binary shift register. The shift direction can be set to be to the right or to the left. The cil, cih, col and coh carry signals provide the chaining functionality. The is a separate signal input whic is XORed with the the carry in signal before it is entered into the shifter. This is to be able to set up loops easily, when the XOR function is not wanted the in input or the carry inputs can be left unconnected. The free / clocked selector in clocked mode will sync the reset input to the trig input, otherwise reset will be asynchroneously. The T/G selector switches the outputs between trigger and gate mode, in trigger mode the output signal will be anded with the trig input, in gate mode outputs stay active for a full input clock period.
And while shifting … I've also made a shifter with built-in XOR feedback (a linear-feedback shift register). The feedback taps to be used are button selectable. Parity switching is provided as well as left / /right shifting. The output is taken from the internal feedback signal, that is the signal going into the shifter. For the output Gate or Trig mode can be selected, in trig mode the output is anded with the trig input. The module can be set to use odd or even parity. An input is provided to allow for external data input, it is XORed with the internal XOR feedback signal.
Also I've added toggle inputs, one for each tap control, to be able to change the taps programmatically.
Made a multiple envelope follower. This one is meant to be used with filter bank modules to create vocoder like effects. Attack and Release time settings are common for all detectors.
No idea what it is good for .. but analog input values an be pushed onto a stack and popped from it, it's named AnalogStack. Also made a digital version of this, just called Stack, that one probably has even less use. The idea was that with a multi sample an hold module and some feedback one can make Mealy and Moore machines. With a push down stack one can make automatons of a slightly more complex class … so … since … as said no idea what it's good for, but they didnt cost much!
A Swan song generator. This module was made after a idea posted on the electro-music.com forum, see the thread at electro-music for more info on it.

The idea is to have a swan, and a couple of hunters chasing it. The swan can fly so it's on an arbitrary position. The hunters though must walk on the paths. The paths themseleves are created by a turtle, for which I made a forth like system to implement it.

This module needs a forth specification file in order for it to work. It will search for mazes.4th in the application directory (which will be included in the next release). This file defines a couple of exports, which are forth words which must not consume nor return any stack entries.

The hunters will close in on the swan on every clock pulse they receive, using the path having the (locally) shortest distance to it, with the twist that no two hunters can be in the same location (when enabled with the 'dups ok' / 'no dups' selector.

Meanwhile the swan can be moved by applying signals to the swan X and y inputs. The output values are taken from the last angle a hunter made and the last distance it traveled (the output being larger when the distance was greater). For inputs and outputs the signal type to be used can be selected. X and Y location outputs for the hunters are available too.

The graph is internally scaled to use an area [ 0 .. 1, 0 .. 1] and angles are scaled [ 0 .. 1]. Stride lengths are used as is, so maze defintions should use a maximum length of 1, assuming that to be the case (holds for all mazes I've made sofar) all input and output values will be in that range (unless being rescaled with the input and output signal type modifiers).

For testing I've made a small program (SwanTester,exe, see below for a screen shot) which can draw then mazes defined in mazes.4th while showing diagnostic messages; I'm using it for designing new mazes. I will include it in the next program release. It will read mazes.4th from it's home directory, so mazes.4th and SwanTester.exe should be in the same folder.

Added an LSystems thingie to make it easier to copy interesting mazes from the webs. Like for instance a Penrose tiling as :
```
  LClear " C" " Clear DefaultLooks"         LMeaning
  " M" " OA++PA----NA[-OA----MA]++"         LRule
  " N" " +OA--PA[---MA--NA]+"               LRule
  " O" " -MA++NA[+++OA++PA]-"               LRule
  " P" " --OA++++MA[+PA++++NA]--NA"         LRule
  " A" " "                                  LRule
  " A" "   1.0 Forward"                     LMeaning
  " +" " -36.0 Rotate"                      LMeaning
  " -" "  36.0 Rotate"                      LMeaning
  " [" " >s"                                LMeaning
  " ]" " s>"                                LMeaning
  " Penrose" " C[N]++[N]++[N]++[N]++[N]" 4  LSystem
          
```
An LSystem will be translated to forth source code which then will be compiled into the forth system. The LMeaning defintions mapping semantics onto the LSystem commands, and the LRule definitions define the system. At the end the LSystem definition specifies a name, an axiom and a recursion depth. The generated forth will be marked with the export directive, making the system visible for the SwanSong module. Any valid forth code is acceptble in the LMeaning definitions.

Added maze type modulation for dynamic maze seletion.
Made a values quantizer module. Mainly to be able to add selective modulation for the SwanSong module, but it seems a handy thing to have anyway.

In the values box a number of comma separated values can be entered. The first and the last value are not being used to generate output values but they define the number range to be used.

This range will be maped to an internal output range of 0 .. 1. Also the mumber of values is counted (minus 2 for the unused last and first values), and this value will be mapped to an internal input range of 0 .. 1.

Now when the internal input value is 0 the first element of the value list (which is the second number in the list) will be selected, and when it is 1 the last element (which is the one but last value in the list), and for input values between 0 and 1 numbers between the first and last will be selected.

The selected number will then be translated to an output value using the determined output scale.

An example : say you enter 0, 0, 1, 3, 4 into the values box, so there are three values (0, 1, 3) and the range of values is 0 .. 4.

The input mapping then is such that input values of 0 .. 1/3 will result in the first value of 0 to be selected, likewise 1/3 .. 2/3 select the value of 1 and 2/3 to 1 will select the last value 3.

Now the selected values will be scaled too, such that the range 0 .. 4 is mapped to a range of 0 .. 1. So the selected value of 0 results in an output of 0/4, 1 yields 1/4 and 3 makes 3/4. The values 2 and 4 are not in the list so 2/4 and 4/4 will never be generated.

This module can be used to select from a number of pre‐defined values. For instance, the SwanSong module has maze type modulation. Now you probably would not want to use all the possible maze types for the selection. For that control you could set the range from 0 to the number of maze types minus 1, and then between those values enter the indices of the ones you want to select from.

Also I've added value indices to the popup menus for selectors to make it easier to select values for this module in case a selector is being modulated from it. These will be put between parenthesis () in the menu.
Made a new delay module with better delay modulation control. A maximum length is set with the range and max length controls. The offset control sets a delay value which can be changed with the attenuated modulation input.
Made a stereo version of the TiltFilter module, TiltFilterStereo.
Added an extra input attenuator on the FreqShifter module to allow for more precise control in the low frequency region.
Added a range selector to the Mux and Demux modules. When for example it is set to four a 'standard' input swing of zero to one will run trough four steps only instead of trough all steps. So basically it is an attenuator on the selection signal.
Added a non-zero option to the Quantizer module. When it is set while the 'ceil' mode is active the output will never become zero, preventing an LFO driven from it (trough it's speed input) to stall. This option is only active in the 'ceil' mode.
Added x-y, y-x and x+y functions to the dyadic module.
Added a speed input to the RandomWalkLfo. Also made the RandomWalkLfo start at a random location after a reset occurred.
Added linear (lm) and exponential FM (fm) to the MultiSine module for each of the sines. These do not have attenuators and operate full scale always.
Added mode modulation to the ChebGen module.
Optimized the Fixed‐ and Prime‐Div modules a bit.
Added scale modulation to the SeqRandom, ProbSequencer and ScaleQuantizer modules.
Added a type III phase detector for the PhaseDetect module. It just multiplies the two incoming signals (double balanced detector).
Added linear FM to the MoogFilter, also made all FM audio rate instead of control rate.
Made an 'auto ready' (auto rdy) mode on the LfoTrig nd SquareLfoTrig modules to allow them to 'self‐trigger' (with a link from the rdy output to the trig input). The default mode (wait rdy) is to only activate the rdy output mode when the LFO cycle executed once, which is compatible with the old time behaviour. With 'auto ready' active and the rdy–trig link present the modules will behave like a regular LFO, exceot that the cycly count is in 'half cycles'. for instance one can make a triangle LFO which starts low goes up, low again and ends up, so it will make a quick jump to low befor the cycle restarts (a triangle with an occasional (depending on the count setting, which is modulatable) saw shape inbetween). Or a delay could be patched between the rdy out and trig in … all kinds of patterns are possible that way.
Changed the displayed values for FM modulation attenuators to reflect the number of semitones of modulation to occur for an input signal swing of zero to one. I.e. when set to a value of 12.0 an input signal of 1.0 will up the frequency by one octave. Also added note names (scientific i.e C4 for middle C) to be displayed in the popup hint window for frequency control knobs. Also I've made a global switch to select note names or frequencies to be displayed in frequency displays (in the Settings Screen F3)

maze tester program
The SwanTester program, for testing the contents of mazes.4th

in version 2017.7.31.0

Consider this release to be a beta version. I've changed quite a lot of internal stuff and have added some half finished new features. But it's about time for a new release, and some bugs were fixed … at least.
Error in Divider module, in the pulse mode it would flash the output too often. Fixed now for the next release.
There was an issue with inserting and deleting 'non-active' modules, like blind panels and the notes module. Such an action would cause the modules in the editor patch and the (compiled) synth patch to get out of sync, resulting in the lights not being updated (and possibly worse). Fixed this to always resync when modules are being added or deleted even when no structural differences between the old and new compiled patch are present.
Pasting text into a display editor (like in the notes module) did not work properly, as windows on double click on a static text control will fuck the clipboard. I've changed the way to call up the editor, must use Ctrl+Left Click now instead of a Double Click. It is also possible to focus the display with a Left Click and then use Enter. Will need to update some help text too … k … diddat.
On display editing EndStateChange was not always executed when editing ended, resulting in the patch no longer being updated correctly on wire edits. Also resulting in strange behaviour on undo and redo operations. Fixed this. Also there was a second mechanism locking undo/redo without performing a recompile at end, possibly resulting in patches no longer being compiled. Removed this mechanism entirely.
Added a speed input to the RandSig module. This is a linear FM input, but for the RandSig module the input value will be clipped to a range [0.0, 1.0] (where a speed of zero will stop the output from changing and a value of 1.0 will make it run twice as fast as it would with an unconnected speed input. So it is not trough‐zero for this mdodule.
Made the LifeSeq module rate smart, so it can run at audio rates now.
Sped up the AD and DA conversions as used in the DAC, ADC and Gate modules a bit by using lookup tables for bit shifts. This also fixed some levels, like the DAC would output twice as much as it should have. This will change existing patches making use of the DAC or ADC. It also changes patches using Gate modules with a bitcount other than one.
Changed some things in the internal working of the code, handling of controls, the way inputs and outputs are being specified, some control flow trough the program. This may result in some stuff not working anymore, sorry about that. I will fix it when I notice it … but having almost 250 mdule types now and various ways different modules can interact with the patch I may have missed some things in testing.
Added a manual compilation mode to the program. When it is on patch changes will not result in recompilation, instead they have to be done manually. When this mode is on you can make changes to the patch without the output going silent, then when done a manual compile (or leaving manual compilation mode) will effectuate all changes at once. The keyboard shortcust for this are Ctrl+F12 to toggle the mode and F12 for manual compilation. The help desk will get cross with me for this one … but, there is some visual feedback on this in the form of a flashing 'manual compilation' text in the top pane and also it is made visible in the Action menu. When there are any uncompiled changes the 'manual compilatin' will flash yellow, and white otherwise. At program startup the feature will always be switched off (it is not saved in the ini flle). Knob changes will still change the patch when manual compilation is on ‐ it only affects rewiring and module insertions / deletions, but this includes loading a new patch or performing any undo / redo operations.

Anyway, thought this might be useful for live patching, we'll see about that. The work flow would be to hit Ctrl+F12 to turn the mode on, make some changes, then hit Ctrl+F12 again to compile the patch and hear the changes.

Actually this is pretty weird on structural changes, it may cause some unexpected behaviour. Still, to do a bunch of knob edits, and have them kick in at the same chosen moment has been useful for me.
I've changed the default behaviour for modules and controls to be enabled for randomization. Randomization used to be all enabled by default, it's now all disabled by default. I've also changed the feedback on this, screen items enabled for randomization are now marked as being special, this used to be the other way around. (Gabe you were right about this from the beginning, took me some time to see it, thanks :-))
I've added some control over the randomization range. It is now possible to set a minimum and a maximum value for randomization on knobs. This can be set by changing the knob value with Ctrl or Shift being held down wile changing the value (either trough the mouse, the mouse wheel, the keyboard or by clicking the up / down arrows under the knob). The range, when not maximal, will be indicated by an arc being painted on the knob's edge (welp … it doesn't look too pretty right now, but it works) and the selected range is also visible in the hint popup (when simple hints are not enabled). Ok, not the best image, but it's the blue hat on the topmost knob … so here is another one, the hat can be red too, meaning the maximum is less than the minimum, giving inverse response (not that such matters much for random signals). It is still possible to navigate the knob outside the set range manually. I did not implement this on selectors … as of yet … As I'm not sure how to give visual feedback on it, and also there is an issue with some of the selectors having dashes in their value list … separators … those are a bit of a PITA. I did implement it for the simple knobs not having a knob at all but jsut the up / down arrows. Here the results will be visible in the popup hint only.
Made a multi‐sine oscillator, which generates six sines with individually settable amplitudes and frequencies individually settable relative to the main tuning. The amplitudes can be modulated by an external signal. The chain input is summed with the oscillators into the output. There are mutes for each sine individually and on the resulting output as well. The sync input when going active resets the phase to zero for all the oscillators. The main tuning has exponential FM modulation and a separate frequency fine tuning. Id did not find room to include PM, but each oscillator has a separate AM modulation input.
Made a text writer module which can be used to write text into a file. I use this to couple a radio stream title to a patch, such that when I change the patch the stream title will be updated automatically. It can hold multiple lines of text, and sending out the next text is controlled from a trigger input. When the module does not get triggered it will statically output the first text line. The text file will only be updated when a new line of text gets selected. This happens at patch load time, patch reset or when a trigger comes in and there is more than one line of text. When the file does not exist it will be created, and when such fails nothing bad should happen.
Made a stereo version of the cubic clipper. It treats both channels independently but with the same parameters, as if there are two clippers set up with the same alpha and hysteresis values.
This module will step trough recorded automation snapshots, this can be complete snapshots or recorded single value changes. The control can loop once it reaches the end, and it can be made to play in two directions.
I've added value history on knobs, modules and the patch as a whole. This with the possebility to undo an edit without a recompilation being needed. It is also possible to make selective undos on the patch by selecting a bunch of individual items in the list. This works on the patch level only, for the control and module undo only one item can be selected from a popup menu. See the images below.

With this I've also split the undo history up into two separate histories. A history for structural changes (those can be undone and redone with Ctrl+Z and Ctrl+Y) and this new 'value history', which can no longer be undone/redone with Ctrl+Z and Ctrl+Y. The control history, on undo, will not silence the patch while a structural undo / redo does need patch recompilation, which will stop audio briefly.

Also this causes automated changes (like randomization, morphing etc.) on the patch to not be recorded in an undo history anymore. Which also causes live morphing to no longer mess up the undo history, it is just not being recorded anymore. (but you can still make snapshots for the automation recorder.)

Still working on a method to let cable changes not cause silence, as those would not really be needed (a cable change results in some pointers being changed only).
And while at it, I've also made an event recorder. In its current state it can only be used to play back snapshots, timed playback is not working yet. With the new AutomationStep module described above this can be controlled. I'm working on more features, and also working on keeping things consistent when modules are added or removed from the patch (it's all a bit messy still, trying ideas mostly). See image below, and the mini version of it on the left.

There is no automatic Save and Load currently, but the automation control screen has a load and a save button. Automation is saved into a separate file with the same name the patch jas file has and an .automation extension, on load it will be checked if it matches the patch it was generated on.
Another thing I'm working on is an abstraction editor. It is possible to make details of a patch public, the abstraction editor will pick up all the publics and generate a module from that with the publics at odd positions, but stuff can be edited. This is even more just in the 'gathering ideas' phase. Anyway, I'll not disable it for a release even whn it is not very useful currently. See image below.

There is no automatic load and save for abstractions, but there is a Load and a Save button in the abstraction editor. Abstractions are saved in a file with the same name as the patch it was generated from using .abstract for the file extension. On load it will be checked if the abstraction matches the currently loaded patch.
Added a 'unique ID' (GUID) to patches to keep track of mappings between patches, automation and abstractions. These should not be to easily tracable to the computer which generated them … when I can believe the µSoft docs on this. Anyways, you can always strip them out, the patch will still load.

The GUID will not change during the lifetime of a patch, not even when structural changes are being made. This means that the sync between patches, automation and abstraction may still get fukt. At least its not possible to load abstractions and automations on a patch it was never based upon.

I'm trying to make it so that when a patch is being change structurally and an abstraction or automation is loaded for it that these will be updated in a sensible manner. (And eventually such would mean automatic save / load will be needed on abstraction and automation files (or tht they get stored into the patch itself).) Anyways … for now, you will loose somme work probably while working with the new features.

Control history
Single items can be reached (and be undone) trough the control popup menu.

Module history
Single items can be reached (and be undone) trough the module popup menu.

Patch history
Multiple items may be selected to set them to the value present in the 'Prev. value' column.

Snapshot recorder
The knob on the highlighted module marked in green is the currently selected control, the one to which the graph applies. In the patch mod 1 is the player, allowing for snapshots to be played in a clocked manner (not respecting the time they were made on). So his acts as a sequencer, basically. A click on the send button will send a whole column of values back into the editor. The Clear button is destructive, it removes all recorded events. Points in the graph can be moved around, when moved in value only the value will change for that step, when moved in time a new time steop will be added. I'm working on more functionality.

Abstraction editor
Items marked in black are made public, the green item is the currently selected on in the editor for which properties can be changed. This is a toy for now to gather ideas.

in version 2017.5.21.0

There is an issue with the Seq16 module in that it will sometimes not run due to the number of steps being seen as zero, I've fixed this. Meanwhile a workaround is to press F7 or F12 to make the sequencer run again.
There was still a race condition in the PulseSync module, solved now. A negative edge on the sync input would clear the internal state, but the incoming pulse might not have been processed yet (in case of a long sync pulse time). Made a counter now which will clear the internal state and the output after a fixed number of cycles, and in case the pulse input had gone active again within that time it will just be armed anew.
Changed the module order on the Control tab to have all scaling modules grouped on the left side.
Upped the maximum count values for the triggerable oscillators and LFOs. This will change existing patches using these modules.
Added Hold mode on EnvAHD module. When on it prevents the module from going into Release as long as the trig input is still active. When off the module works as it always did.
Added an active output to the LfoTrig modules. This differs from the inverse of the ready output in that ready will never be active before the first cycle ran (and the module will not run then yet when it saw no trigger signal).
Added some swell control to the EnvAHD module. Swell when active will cause a rise during the hold phase of the envelope. The amount of the effect can be preset and modulated, and the shape can be selected. The time for the swell is the current hold time. The envelope graph more or less shows the effect of this new control.

Added swell to the EnvADSR module as well. It works a bit different here, the swell time is taken from the release time instead of the decay time (as there is no hold time to be set). Also setting a large amount of swell will reduce the attack level by up to one third.

With swell set to zero the modules behave as they did before. For shape impressions see the image below.
Added a freq (unattenuated FM) input to the HrastOsc, also added shape and morph modulation on it.
I've changed the sensitivity of all speed inputs on LFOs and oscillators. They will run twice as fast as normal now with an input signal of 1. A zero input signal will make them stop, and with negative values they will run backwards. Of course the only wave shape changing on that will be the saw wave. This will change the behavior of some existing patches.
The phase modulation on oscillators and LFOs was changed to something more correct (for large phase modulation values), turned out to be a tad cheaper as well.
For the following modules I've changed the range of the slew control to make it more useful: TypeFlip, SeqChord, Slew and SeqRandVal. This will change existing patches making use of the slew control on these modules, however when slew was off on them it will still be off for the new control range. Also it is still possible to fully freeze changes with the control turned hard right.
Added 'function modulation' to the Gate module. The blue input when going from zero to one will go trough the eight selectable logic functions (it is being used as an offset (modulo 8) on the value from the Gate type selector).
Renamed the Pulser module into Wavolver ‐ as it was really that … and then made it work properly. The renaming will break existing patches using this module, the module will still be loaded but all connections to it will be lost. But then again, the name was plainly wrong, and it did not function properly either. The actual pulser thing can be done now with the triggered oscillators, I've added pulse count modulation to those. These are all variations on VOSIM of course.
Added a new display mode to the Quantizer module. It used to show fractions of notes, it can now show fractions too. Not that this changes anything on it's workings, but with the mode set to Fractions it is easier to understand how it works for linear control of (low frequency) oscillators. For instance (see illustration), when the fraction is 1/4 and the rounding type set to 'round' and the output is used for speed control of an LFO the LFO will run at n/4 of it's set speed (for n = 0 to 4 assuming an input ranges of 0 to 1, and assuming the linear frequency control is being used (called speed in most places)). I've added the same option for the built-in quantizer in the SeqRandVal module.
I've added different rounding types for the internal calculation. The module did use the Round method before, but it can now also use Flooring and Ceiling. The default still being Rounding. The illustration shows the quantizer set to 1/4 with from top to bottom the input signal, the floored quantized value, the rounded one and the one using ceiling. This option has not been added for the SeqRandVal module, that one will always round.
Changed the Song module to have active 'end' and 'silence' outputs at patch startup and after reset (both global and module specific), they used to be inactive then. This makes it easier to use a PulseSync module on the 'end' and 'silence' outputs. This may break some patches that are making use of these outputs, but not likely so.
Added a signal shaping module, it can transform a linear input signal into an envelope shape type output signal with inversions and signal types being applied. The module is rate smart, so it can be used for audio signals as well (but there will be some aliasing). For shape impressions see the graph below.
Added yet another scaling module, the PropScaler. This one has a modulatable parameter p which controls the in to out transfer as:
out = in * ( p - 1) + p.
This replaces a set of four modules that would otherwise be needed to make the same functionality. Effectively p sets a lower limit on the output value and the input range is proportionally distributed over the remaining output range.
Added a new sine oscillator with more modulation control than the other oscillators have. It has amplitude modulation, phase modulation, linear FM modulation (trough zero), exponential FM modulation and in addition the frequency fine tune can be modulated trough the ct input. The sync input is a hard sync. when it goes active the internal Phase accumulator is reset to zero.
This is after an idea by Ian Fritz. The analog xor makes it's output active only when the sum of the input signals is over the set set Threshold value, but below 0.5. The threshold can be set from -1 to +0.5, and it can be modulated trough the T mod input.

Envelope shape impressions

Shaper shape impessions

in version 2017.4.22.0

Moogfilter could produce denormals - added some guards there. Also optimized the filter a bit (20% off from the execution time), but it still uses an awful lot of cycles ‐ more than a reverb does ‐ I don't really understand why that is.
Compressor could produce denormals - added some guards there. Pluck module idem.
Graph Image for S shaped attack on the ADSR envelope module was wrong, fixed it.
Made the S attack and decay modes working on the Seq16 module, they acted like exp by mistake.
Added code for trying to get some better idea as to why PortAudio initialization failed. It will report failures to the logs.
Added an off-beat output to the EuclideanRhythm module, it is active whenever the regular output is not (and the clock is active). Basically it runs the reverse pattern of the regular output.
Added an invert input to the Seq16 and the Sequencer modules. In invert mode the sequence is put upside down.
Added inputs to the Switches4 module. When these are not connected it will behave as it used to. And when they are connected they will be switched to the outputs.
Added range modulation to NoteQuantizer, ScaleQuantizer and ModQuantizer modules.
Added tanh( x) function to the Monadic module.
Added chain inputs to the TransInv module, these are just added to the calculated output signal.
Made the pulse delay module rate smart ‐ however ‐ the shown times are eight times as short when the module runs at audio rates (this is being indicated in the displays). Also made the MultiCompare module rate smart, so that now all the logic modules can switch to audio rate computation.
Added some values to the MConst module to make it usable for controlling note intervals. uni (0), mi 2nd (1), 2nd (2), mi 3rd (3), ma 3rd (4), 4th (5), dim 5th (6), 5th (7), mi 6th (8), ma 6th (9), mi 7th (10), ma 7th (11) and octave (12). These values are TET based and are fixed, even when the main tuning is changed.
Added a sent output to the MIDI CC output module to allow for easier sequence control.
Added a clear input to the Delay module, when it goes from inactive to active it sets all the delay memory to zero. This is glitchy, as it was intended to be.
Added input attenuation controls for the Formant and Formant2 modules.
Added a speed input for Osc, Square, MultiOsc, OscTrig and Squaretrig oscillators. This basically is linear thru zero frequency modulation. When speed is set to zero these oscillators will stop and they can be used as lookup tables by controlling them from the pm input. And while at it … also added speed controls for Lfo, SquareLfo, LfoTrig and SquareLfoTrig modules.
Changed the maximum acceptable input level for the GVerb module to 100, in the original code this was 1e5 or so and had that lowered to 10 before ‐ but it could cause distortions that way … so upped it a bit. These are silly levels, but this can happen when the input level is high and the output levels are set low to compensate for that. Which would be silly to do probably, but I needed some more forgiveness there.
Added a 'hold' mode for the EnvAR, EnvARRetrig and EnvConrol modules, when active they will not enter the release phase as long as the trig input is still active. Also removed a possible click from these modules ( on the transition from Attack to Release or Hold mode) and removed some needless calculations … like subtracting Ln( 1).
Added mode selectors for the FixedDiv and PrimeDiv modules, these can be set to either Square or Pulse. In Square mode the outputs will all have a 50% duty cycle and in pulse mode they wil be active for one positive input clock time.
Added a reversed mode for the arpeggiator, and added a reverse input for playing the arpeggio backwards. The reversed mode plays the notes in reversed order of reception i.e. last note first. Also added a transparent mode, when the Transparent control is active no arpeggiator actions takes place but instead the last sampled input value is being output when the output clock goes active (making the module behave like a sort of a sample and hold module (except that it has two sample clocks then).
Added a trans(parent) input on the Poetry module. When it is active no poetry generation takes place but instead the last sampled input value is being output when the output clock goes active (making the module behave like a sort of a sample and hold module (except that it has two sample clocks then).
Added count modulation inputs for LfoTrig and SquareLfoTrig modules.
Added a level modulation input for the Quantize module.
Made the control decimation programmable, it can be set from the Settings Screen F3. The default value is 8, what it always had been, but it can be lower or higher now. Higher values will mean that less processor resources are being used. Changing the value however also means that some modules will run faster or slower, this affects some of the blue filter modules i.e. the slew times may change etc. It is of no concern for the red (audio) filters though.
Working on some MANIAC features …
I've dded cross fade modulation to the Seq16 module and also added a random input. The random input will set all the faders to random positions. Added three 'canon' outputs, these are shifted by four steps for each one. So when the first output outputs step1 the 2nd out sends step 5, etc. This way the sequencer can be used as a 2 x 8, 3 x 6, 4 x 4, … sequencer, but it is also possible for instance to set like 12 steps and use output 1 and 3 and then they will have eight sliders overlapped and four unique ones for for themselves. The xfade works on all four outputs, but there is still just one chain input which is being added to all four outputs. Added a 'hold' input, which when active will hold the last sequencer output value and it will also block the chain outputs from being updated, but it keeps the sequencer running. Then added skip inputs for each step, when a skip input is active that step will not be used (making the sequence a step shorter). Added a clockskip input too, when it is active the sequencer will freeze at its current step. This is different from the Active input in that the Active input outputs no step values but passes on the chain outputs only. When a clock is being skipped the current step and output values will be he held and the chain inputs are no longer sampled. Then I went on and made 16 step outputs, going active when the step is active and the incoming clock is high or the trigger mode is set to (G)ate. Oh, and also added a TimeWarp (tw) output, which can be used to control the speed of the driving clock, it is meant to drive the speed input of an LFO. The idea is that when steps are being skipped the sequence length can be kept the same when using the TimeWarp output in this manner.

Getting closer to the M.A.N.I.A.C.
Added the 'canon' outputs on the LUT module as well. Added a 'steps' control as well to map a zero to one signal range on the set number of steps, step count can also be modulated. Then added a random input for doing fader shaking too. With a negative step count (which can be set trough the steps modulation input) the unit will run 'backwards' ‐ that is ‐ for larger selection values it will move to a lower step value. Note that with fewer steps and a ramp on the input there will be less steps in the same time, so when used as a sequencer the sequence will run slower then. Added 16 step outputs here too, there is no clock or active input on this module, so the step outs are active for the whole time the step is active.
Added a DC blocking module with a -3dB point that can be set from approximately 20 ~ 40 Hz to 400 ~ 800 Hz in audio rate (red, depending on sample rate) mode and from 2.5 ~ 5 Hz to 50 ~ 100 Hz in control rate (blue, depending on sample rate) mode. The further the knob is turned right the higher the cutoff frequency will be. DC will always be totally blocked no matter what value the cutoff is set to. Feedback delay lines (and there are a couple of other modules that do not like DC very much) can lock up when there is DC present in the loop, this module can be used to get rid of the DC, probably best before the loop and not inside it.
Added a MIDI Program Change Output module, also featuring a sent output for easy command sequencing. It will send out a a program change message when the gate goes active or when the program is manually changed. The actual program is the SUM of the values read from the prg control and the progr input. A change on the progr input will not trigger a message to be sent ‐ a gate signal is needed then.
Added a MIDI Parameter Number Output module. This module can send both RPN and NRPN data and it can do so with the most significant septet only or it can send both data septets. Added the chain output to this one as well. Data will only be sent when the gate signal goes active.
Changed the compressor looks a bit. The LED scale was not very informative (as the scale is pretty large things barely moved, and there was no feedback on actual gain either) so … removed it, but added a dB value indicator showing the current amplification.

Also added a side chain input and output. Effectively this is a 3rd compression channel where the compression is being calculated on either the side chain channel (side mode) or on the stereo signal channel pair (norm(al) mode). So it is also possible to side-chain a mono signal from the stereo inputs.

Then added a DC blocker on the inputs as in RMS mode DC on the inputs can do some weird stuff. The blocking can be switched between AC and DC. AC mode has a -3dB cutoff point at approximately 20 ~ 40Hz (depending on the sample rate) and DC mode does nothing, The filter will only operate in the compression calculation chain, the signal chain is not being filtered.
Added a PulseSync module. The module gets triggered by an incoming positive edge on the pulse input, which it will then hold internally. When a subsequent positive edge appears on the sync input the output will be set active. Once the pulse got processed the output and the internal state are both being cleared when a negative edge is seen on the sync input. The incoming pulse is guaranteed to never get lost.

This module can be used to sync up a pulse event appearing at a random time to a syncing clock signal.
Added a Transpose module. Four separate transposition amounts can be set each having an activation input. The output will be the input note value transposed over the sum of the active transpositions. Transposition range is -24 to +24 semitones.
Added a SquareSineLfo module, it outputs a sine along with a PWM controlled square wave. I often end up clocking things with sine LFOs as sine are easy to multiply up to higher frequencies (using the ChebGen module) which is easier often than starting with a higher clock and dividing it down. Also the sine shape is more useful to do (synced) modulations with. However its not easy to change the 'pulse width' of the sine, so that is what the square PWM is for, that one is easy to make changing note lengths for driving sequencer.
Made a PulseSkip module. It will pass on input trigger signals only when the input level is larger than the compare level. Levels are computed as the sum of the preset level (as made visible in the displays) and the modulation level. Once a pulse got started will only be ended when the trig input goes inactive, so either a complete pulse is being passed on or it is being skipped.

in version 2017.3.14.0

Bug in DataGraph module for the splines mode, it outputs X values instead of Y values. Using Y works way better - so that is fixed now. And its not in the AudioGraph module, as that one can not use splines ‐ pwew :-)
Mode and range changes through selectors were not handled when controlled with the keyboard (only mouse clicks worked for that), fixed it.
Added an 'act' input for the EuclideanRhythm, SeqRandom, SeqRandVal, Pattern and ProbSequencer modules.
Added inv outputs on EnvAHD and EnvADSR modules, they output 1.0 minus the value that is on the env output.
Made a controller module for the DataGraph module to allow it to do envelopes. The help blurp says : A retriggerable envelope controller. The envelope will start when the trig input goes from zero, or below zero, to above zero. The output will then in the set Attack time go to the level set with the level control. When the trig input falls off the output will start getting larger again for the set Release time, till a value of 1 is reached. At every new trig start the output value is reset to 0, and the above cycle will restart. Both the Attack and Decay phase are linear, any non linearity is to be set in the controlled graph. This module is meant to control the GraphControl module, the actual envelope shape is then drawn in that module. The set level will then correspond to the vertical distance in the graph where the hold value will be.
Made a pattern sequencer module to be used for drum patterns. When set at low speed it can also be used to control the time structure for a self playing patch.

A full pattern is made from eight parts, where each part is composed of eight sections and each section has eight elements. This makes for a total of 512 steps, where each step controls 24 bit outputs simultaneously. The steps can be set to off, to pass the incoming clock, or to be on.

For detail information please click on the module, or for some background information have a look at the forum thread.

Sequences can be saved to disk and reloaded just like any module (or any group of modules) can - as a preset. Right click on the module and select load / save preset from the context. menu.

in version 2017.2.25.0

Range control for the pulses module was not working, fixed that.
Output signal type selection for the pulse delay module was not working, fixed that. Also tried to clarify the help text a bit.
Constant module when put into 'note' mode is off. And this seems to be wrong in more module types. Indeed it did affect the Constant, LUT and Knobs4 modules and it is fixed now. Also made it so that when 'notes' mode is selected the note controls have 128 steps now, and possibly some other number in the other modes. Also the LUT module was having an offset in the 'notes' mode, this got fixed too. These changes will affect old patches using any of these modules in 'notes' mode.
Made a Moog style 24 dB low pass filter. Inspiration came from a github site, the Huovilainen variation on the theme. I've found it to be a bit explosive … the volume that is, so made it to oversample a bit more and also added a saturation function (tanh) on the output, as it had a tendency to go like very loud when resonating. This is not a cheap filter CPU wise, but I think it to be a nice one with the built-in distortions. On the referenced web site there are some more variations, I may try some other and make a selector maybe for different types.
Made a fixed divider module. The module has a series of outputs with fixed division rates. After reset all the outputs will be off, to be toggled each time the number of input clock periods matches the division count for that output. The output will stay high for one clock period then.
Made a multiple waveform oscillator. This module simultaneously makes a sine-, a trianglar-, a saw- and a square-waveform. The PWM input provides linear Pulse Width Modulation (on the square output only).

in version 2017.2.19.0

Delaytime range on stereo looper was not working, fixed now. Also changed the start and length control modulation amount knobs to be displayed as linear (they were actually working as linear controls already but not displayed as such).
Some screen painting took place on the desktop's device context, changed that to happen on the editor's context only. This draws better selection rectangles and better lines (on cable drawing). This may as well solve an issue with the screen not being repainted properly (which was reported by someone using Windows 10). Also thickened up the lines a bit for better visibility.
Added inv outputs on EnvAR and EnvARRetrig modules, they output 1.0 minus the value that is on the env output.
MIDI transmit bug. The command was never sent only the data bytes … Fixed now. And of course I had forgotten to test some code …
Added f(x) = - x and f(x) = 1 - x functions to the Monadic module.
Added a a modulation depth control module . The graph illustrates it's workings (where the red signal is the wanted modulation depth (mod), the green signal is the modulation (in) and the blue signal represents the resulting output). For the inputs and the output the signal type can be set, for the graph they were set as in the module image. The mod signal and the output being unipolar and the modulation (in) signal being bipolar. I've made this module primarily for AM modulation (tremolo), but it would work for vibrato as well when applied to a PM input of an oscillator.

The internal calculation used is : out = mod * ( in - 1) + 1, so that when mod = 0 the output will be 1 and when mod = 1 the output will be a copy of the in signal. This is being calculated on signals assumed to be positive only (unipolar), the input output type selectors are to be used to ensure such scaling.

in version 2017.1.29.0

When windows screen magnification is set to something else than 100% everything on screen gets messed up. Turned off screen scaling, which paradoxically results in better scaling - I guess the scaling property on forms is a promise to windows ... anyways, windows will now scale all stuff properly - or so it seems.
Bug in LifeSeq module, the chain input does not work ‐ fixed now.
Obscure bug in the wave file reader code, it would read more samples than present in the audio file which would cause glitches on playback. Fixed now.
Some of the 6 dB filters (the high pass parts) could produce denormals, fixed that.
Bug in multiplier module where the leftmost input would not work due to having a wrong name. Fixed that, but all patches using that input were not working properly and on reload will not restore the connection. When needed this can be fixed manually. Open the patch in a text editor, search for all occurences of the text 'multiplier_in47' and replace all of them by 'multiplier_in7', the patch would then load again as intended.
Bug on pasting modules. It looks like when a selection larger than what is present in the current patch can not be pasted. This has to do with an internal naming conflict for modules. This issue must have been present for years I think, looking into it now. Is fixed now, it made the code somewhat simpler too :-)
Added a reverse input to the LifeSeq module, when active it plays from right to left. Also added stops to the module. When one stop is set it will play from the first column up to the row where the stop is set and it will then jump back. When two stops are set it will start in the column after the leftmost and it ends in the column just before the rightmost. When the reverse input is active this gets reversed. When no stops are selected the full range will be used. On reset the cursor will jump to the column after the leftmost stop, unless only one stop is set ‐ in that case it will go to the first column. This also gets reversed when reverse mode is active. I thinks the stops to be interesting mainly for the sequencer mode.
Changed MIDI NoteOn handling a bit. When a NoteOn with velocity 0 is received it gets translated into a NoteOff event. The velocity for the generated NoteOff was set to zero, it is now set to 64 instead. Changed that for the MIDI file player module too.

Implemented running status for the MIDI transmitter, this was already present for the receiver.

Updated the MIDI decoder and encoder to be able to handle RPN and NRPN data, so could make modules handling those MIDI messages now. Implemented a mode for this. When 7 bits mode is on the callback will be called when the MSB comes in, in 14 bit mode the callback will be delayed until the LSB comes in. When MSB and LSB get swapped by the sender it will still work, provided 7 bit mode is selected, as in 7 bit mode LSB reception will not trigger the callback (and the internal value gets updated anyway on both MSB and LSB reception.

Data increment and decrement are implemented as well, in 7 bit mode they will change with steps of size 128 and in 14 bit mode with steps of size 1. Increments and decrements are clipped to the range [0 .. 16383] (which may get interpreted as a range [-8192 .. 8191] depending on the actual receiver), no wrapping takes place ‐ could make a mode for that too though I guess.

All this is in preparation for MPE (Multidimensional Polyphonic Expression) … going to be some messy fun that for what's basically a mono synth … sorry paraphonic :-)

But first MIDI will need little more attention.
[partially obsolete] For PortAudio I've changed the way buffering is handled. It is now possible to set buffer size of zero, whech means that PortAudio decides on it. This works for ASIO as well.

[As of 2019-04-10 legacy audio mode was completely removed.]

For legacy mode this is not possible, a 4096 byte buffer will be used in legacy mode when 0 was set. I've noticed that the WASAPI interface gives me best results for latency, but I've also noticed that I'll have to shave off some of the peak load for some of the modules to be able to use that for larger patches. I've added a buffersize display under the CPU load display to be able to see what gets actually being used (it shows the number of samples delivered and requested by the PortAudio callback). Setting the buffer size to zero may cause the VU meters of the main window to be updated a tad irregular, but didn't actually really see that. Also, in the Debug Screen F5 I've added latency indicators, they show what is being reported by PortAudio.
Processing for the LifeSeq module is spread a bit better over time now. After a change request was seen a column counter is set up which gets adjusted on every control rate tick. On each tick only one column of the automaton gets updated. As a consequence the updates should not occur any faster than once every 32 control rate ticks (SampleRate / ( 8 * 32) Hz). Also the display will now get updated only when there was a change.
Made it so that all MIDI sends from a patch on a channel > 16 are sent back into the patch itself (and the channel is decremented by 16 then). This makes it possible for a patch to send MIDI data to itself by allowing it to send on a channel > 16. For the MIDI send modules these channels can be selected as ic 1 to ic 16 (ic standing for Internal Channel). The MidiNoteOut and MidiCCOut modules now allow for such channels to be selected, so a patch can now send notes and CC messages to itself. Such MIDI messages will be handled internally in the synth patch and will not be visible outside it (like in the user interface, or in the outgoing MIDI stream).
Two new modules, a Phasor and a painted waveform thingie, the AudioGraph module

The Phasor was created to control the AudioGraph, but it can also do phase control for oscillators.

The AudioGraph is better suitable than the DataGraph module for audio rates as it internally uses Differentiated Parabolic Waves to reduce the aliasing a bit. I've added an auto scaling feature which after every data alteration rescales the graph to use the full Y value range. This feature can be turned on or off.

Also added an input to randomize the graph, randomization will occur when the random input goes from zero or lower to something greater than zero. The actual momentary value at that stage is used to control the amount of random to be applied. Small amounts (near zero) of random cause the graph to be changed just a bit, a value of one would create a totally uncorrelated new graph. Unfortunately the rnd input acts sluggish … maybe something to fix later on.
Added a tilt filter from the music dsp site. A center frequency can be set for it and a tilt factor for the frequency response line. With tilt in the mid position there will be no filter action, turn it up for more highs or down for more lows in the output signal. The center frequency is the point the response line rotates around. The maximum tilt is ± 6dB / Octave.
Added a txt end output to the Song module, it will fire when all phonemes are spoken. It will not fire initially so it can not be used to make the module self looping. However, the module does have a loop mode for that purpose. I've added it to be able to only start the module when it was not already running (with an RS flip-flop added).
Added hold mode to the Demux module. When active XFade will no longer work and outputs will keep their value when they get muxed out (integrated sample and hold). The new button toggles between cross fade and hold modes.
Added a sync output to the RandomWalkLfo module. The sync output goes active when the output values change and it goes inactive halfway the LFO period. This so I can get triggers at the LFO rate. Also added an output smoothing mode to this module. When smoothing is off a stepped wave is produced, when it is on the wave glides from old to new values. And also added a distance modulation input.
Added a maximum value display to the Value module, it will display the maximum of the absolute value seen. A reset button was added too to be able to reset the maximum to zero. Also a patch reset will set the maximum value to zero.
Changed ini file loading a bit, at first Wren will try to load an inifile based on the name of the actual executable file name. When such does not succeed it will try to read a file 'wren.ini' from the same folder the executable is in. This assumes Wren originally will be called 'wren.exe', but makes it possible to have a copy of the executable under a different name which will then try to read the original settings (but they will be saved into an ini file named after the actual executable fiel name). Anyway, this should make it a bit easier to have two (differently named) Wren executables in the same folder, and to have the new one to initially inherit the settings of the default one (they can be set to differ later on). The looks will always be shared.

in version 2016.10.23.0

Oops ... a bug in the AM modulator causing an addressing error, fixed now.
I've changed the implementation for the intdif module a bit. For the integration function the output value is divided now by the execution rate of the module. This can be scaled with the gain control slider, but it means the output will be way lower than it used to be. Likewise for the differentiation function the output value is multiplied by the execution rate, and this is also scaled with the gain slider. This results in way more output signal for the differentiator. Also the output values for both integrator and differentiator are clipped to a range [-10 .. 10] (this used to be the case for the integrator only). The execution rate depends on the set sample rate and it can be either control rate or audio rate depending on the rate of the input signals.
Trying my hands on some Markov stuff. Not sure if I can get the learning mode fast enough for real time use, playback should not be a problem.

This is a first order implementation.

The internal state consists of a number of nodes each holding some value (for which a maximum must be set with the max size control), a State variable holding the current node index (which is not under direct user control) and a matrix of transition chances (which is not under direct user control).

When the learn input goes active the current input value is looked up in the nodes, and when found the transition probability from the previous input value to this new one will be increased with the current strength value (which can be negative). When the new value was not found it will first be appended to the node list. When the node list becomes full and the mode is set to 'once' an overflow condition occurs and the new transition will not be learned (also the Ovfl light will go on then). When the mode is set to 'cont' no overflows will occur but the oldest transitions will be discarded instead.

When the 'new output' input goes active the current State variable will move to a new one, this is a random process controlled by the transition chance matrix.

The state of the module can be saved to and loaded from a file (must still implement some user interface for that), so … training could be done in non-real time mode when the need arises. Anyways .. let's test things first ‐ actually it does not seem to perform so bad, great :-) It will depends on the input material, when there are many different values things may be worse, but for discrete events (like note values) it seems all fine.

Now it needs to be made a bit leaky … as currently when the input patterns change it will take longer and longer before those new patterns will be learned. Although setting random strength values seems to help here ‐ just like setting not too many nodes and using 'cont' mode.

Now working on some code to do do the Markov thing a bit more efficient so I can hopefully implement a second order implementation which is fast enough for real-time learning.

Ok, that seems to work. It also seems to be not too hard to generalize it a bit to arbitrary orders. I've implemented something, but found no time yet the check the workings or the performance.

After a little break and some bug fixes … arbitrary order seems to basically work, it needs more testing though.
Made a variation of he Average filter with direct frequency control (instead of just exposing alpha). The set frequency references the - 3dB point. Filter falloff is 6 dB / Octave, just as for the Average filter. The filter should work on control and audio rate signals (Rate Smart).

Made a bandpass variation on this and then made a stereo version of that one too. I've used this one on the final output of a patch to trim off excessive lows and highs a bit.

When in control rate mode the frequencies are off by the control rate decimation factor (currently 8), the displayed value being too high. I should fix that by displaying the lower frequencies, but as of yet there is no proper mechanism for that.
Added a new tab for generators (gen) and moved some of the 'sequencers' there. This tab is meant to hold modules to be primarily used for generative music. So state machines, song generators, cellular automata etc. Moved some modules from the control tab to the new tab as well.
Automata … well … just one automaton for now, but will likely make a couple more based on the same underlying engine. This one is a rhythm generator only.

This module implements an 'elementary cellular automaton'. Each cell can have a value of either zero or one. Three cells are evaluated according to the set rule and this will yield a new value for a next generation cell. The rule number [0 .. 255] is being interpreted as a bit pattern which is indexed by the three bit pattern generated from the three evaluated cells. This is all according to Wolfram's theory about elementary cellular automata. The start state of the automaton consists of one single cell having a value of one.

With the steps parameter you can set the number generations to be calculated, this results in a bit string. On every step a next bit will be sent to the output and when the end of the string is reached the sequence will wrap back to the beginning.

The Faults parameter sets the average number of cell faults generated for each new cell calculated. The set number is divided by 10,000.

The total length of the generated pattern increases with two on each new evaluated step, which results in length = 2 * steps + 1, which is odd always. The reset input can be used to wrap back sooner.

The blue output has a weighted sum of the last three bits ( Bit[ anIndex] + ( Bit[ anIndex - 1] shl 1) + ( Bit[ anIndex - 2] shl 2)) / 8.0. The dir input when active reverses the play direction.

Some rules will not generate any output at all (like rule 0 or 255), some will generate simple repeating patterns and others will be almost random. For some rules the steps setting may not change the sequence much, for others a next step may differ wildly from the previous one.

For some ideas about possibly interesting rules have a look at my rules page. At some stage I may build some of the visuals into Wren , but not at this stage. That page, at the very end, also shows an implementation of the algorithm being used.
And another cellular automaton based module. This one uses Conway's game of life. the current help text says (subject to change though):

A Life based sequencer. Based on Conway's game of Life, but there are some extra modes selectable from the mode selector. The step input moves the cursor, which will put the cell values from the current column on the green outputs. The change input triggers a new generation of live cells, according to the currently selected rule. The blue sum output has the column sum of the cells under the cursor, this can be a binary sum or a linear one. The grid can be set to be wrapped or unwrapped. In wrapped mode the top connects to the bottom and the left side to the right. The fault rate parameter determines the amount of faults made in applying the evolution rule. When the number of cells drops below the low count the grid will be filed with random values, the chance of a cell becoming alive is set with the rnd amt parameter. The rnd button will fill the grid with a random pattern, obeying the rnd amt setting. The clr button will clear the grid, this usually will then immediatly set random pattern as the live cell count drops below the minimum. It is possible to draw on the grid, a single click will toggle the cell state, with shift cells can be turned on and using ctrl instead will turn cells off.

Using a right click on the grid the current state can be saved to or loaded from a file.

This a a pretty big module, and 32 rows seems a bit much maybe, but I've found the game of life to be more interesting on larger grids. The current 32 by 32 size seemed a minimum for interesting patterns to occur. And at the same time 32 logic outputs seems like overkill as well, for the sum output it's a nice amount though (2.5 octaves in the note modes).

Made an inhibit control on the change input to be able to stop rule evaluation, and also made a 'Freeze' mode (in addition to the various ;ife modes) in which the current pattern will not change ‐ this allows for painting sequences.

Now if only I could get the module to load itself from a patch file … it would then double as a 32 x 32 (event) sequencer (the sum output can be set to make note values though) … it also reinitializes on every module insertion or deletion … which it should not do. But otherwise it seems to work well. Ok, the save and load stuff works too now, but the mode has to be set to 'Freeze' for this to work properly (or else the internal Life machine will have it's own ideas about the pattern).

The new grid control thingie could be nice for other modules as well :-)
When Wren is closed the current patch will be saved in the application's home directory (where wren.ex resides) as '__l__a__t__e__s__t__.wren'. This file can be reloaded from the file menu trough the 'Load last patch' item. For just in case … This is not saved in the most recently used patch list (unless you manually open it; it will not be opened automatically on Wren startup).

in version 2016.10.2.0

Added the Markov module, but it has not been fully tested yet, see remarks in the 'working on' section above. Also I did not implement the code for saving and loading models yet ‐ or rather ‐ I did, and then the code changed, so it needs to be done anew.
Delay module, in 'short' mode, does not get correctly initialized on patch load, it seems fine on patch recompilation though.

Or actually … several parameters were not updated in the synth after for instance a range change on a mixer or when clicking an rnd button on a sequencer, etc. etc. Fixed that, it was introduced with the randomization features in the last couple of previous releases.
Made some new noise modules with Perlin like noise, in which the roughness can be controlled and the pattern length. A clocked one and a free running one. New noise values can be inserted to change the pattern. The maximum pattern length is 4096, the length controls set a fraction of that. The number of layers can be set, more layers give more smoothness, at the expense of processing time. The new input when unconnected is active, giving noise like behavior, when it is set inactive the behavior is tone like. When made active less frequently it will change the timbre. I'll probably make LFO versions for this too, welp, would like to at this time, it seems like a slowly clocked one can make nice percussion patterns. Also I'd like modulation of the length ‐ ok, added that. Not sure how strictly Perlin this is, I've tweaked the interpolation function from one layer to the next to more or less give the same amplitude for smooth and rough settinging (but it is random of course …).
With the above … there now are 200 module types.
And then when on the Perlin you do not set new values very often it makes sense to have frequency control as well. The maximum sequence length was set a bit shorter for this one (256 vs 4096 for the Perlin noise modules). This can make some nice metallic or tone like noise sounds. I should add a button to the Perlins to invert the new input. As it is now the new input is set by default, this is not always handy.

I've added the invert option ( i(nvert) vs n(ormal), the latter being the default value on the noise modules, the former for the tone module).
And another new module, Random scales … the help blurp says something like:

This module implements a scale randomizer. Each time the rnd input goes from inactive to active, or when the rnd button is clicked, a new random scale mapping will be computed. A scale will always have 12 notes per octave, and all octave mappings will be the same (with an octave shift, of course). Within an octave the 12 notes will be randomly mapped to 12 (probably, but not guaranteed so, it's random after all) different notes.
A simple arpeggiator

On each clock on the 'new in' input a new input sample is is taken and added to a list of samples, the oldest value present will be dropped from the list then.

This list can be left unsorted, it can be randomized or be set to play the lowest or highest note first .

The list length is set with the length control.

On each output clock on the 'new out' input the next stored value from the sorted list will be output.

In random mode the order in the sorted list will be randomized when the playout wraps back to the first value.

In unsorted mode the playout will be the recording order.

On the input side an acceptance filter can be applied. All incoming values can be accepted, values differing from the previous value or values not present yet in the list of samples.

The toggle input when inactive causes the selected sort order to be in effect and when active the next sort order will be used. This can be used to toggle between low to high and high to low modes, for instance.

When there are not enough input values available (yet) arbitrary values will be clocked out until the playout counter wraps back to zero.

The reset input (or a global reset) will cause the playout pointer to be set back to zero without changing any stored values.
And another new one, a probabilistic sequencer.

For each note the probability with which it will be played can be set with a slider (and this value can be multiplied by the value on the associated input). Notes have to be be enabled before they can sound, and note enables can be selected from a scale preset.

The presets will be dynamically loaded from a file wren.scale that must be in the same folder as the Wren executable file. When this file does not exist Wren will create it. After that you can edit it with a text editor (like Notepad for instance) and Wren will not touch it anymore.

A selected scale can be altered with the note buttons, and such a change will be saved with the patch (making the shown scale preset name meaningless, of course).

The rnd button will set a random probability distribution, the clr button will set a flat distribution. The distribution can be altered with the tilt control. A zero value means that there is no tilt, a positive value makes higher notes more likely to be selected and negative values will produce more low notes.

A transposition can be applied to the selected scale with the trans(pose) control. C is the 'neutral' base note (not leading to any changes). Transpositions are not reflected in the shown note pattern. The low and high note controls limit the range of possible notes.

The sm(oo)th controls determine the likelihood that the next random value is about the same as the current one. So with a larger smoothness the results will be less jumpy.
Added noise type selection to various noise modules (OSC and LFO). The default type is Linear (which was the only available mode before), the added types are Gaussian (with a fixed sigma of 1.0) and exponential (with a fixed lambda of 3.0). This affects the Noise, NoiseTrig, NoiseLfo and NoiseLfoTrig modules, but backwards compatibility was maintained.
Added mode control to the Pattern module. A mode can be set on each step, it can be normal (nrm), random (rn1), weighted random (rn2, or rn3) or skip. Normal meaning that it will do it's full count. The random modes will generate a random number [0,Count] and use that as the step count. The first random mode considers all numbers in the interval, the second random mode only divisions by 2, and likewise rn3 uses divisions by 3. Skip will just skip the step. Also added individual step outputs which will go active when the associated step causes an output pulse to be made. The step outs will produce a 'per step' copy of the main output.
Added clear inputs to arpeggiator and poetry modules to allow them to forget stuff. These are latching inputs capturing a low to high transition. When they are in the set state a next value to be clocked in will not be the input value but zero instead. After having been processed the set state will be cleared.
Added an Amplitude / Ring Modulator module, which is a multiplier with control over modulation depth and modulation offset. This makes it a bit easier to apply less than full depth AM or RM to a signal. The input signal type can be set, and it can do a fade between amplitude modulation (AM) and ring modulation (RM).
And made a simple pattern rewrite module for which a couple of rewrite rules can be specified, and an expansion level. There is a small alphabet of the letters A, B, C, D and E. For each letter a substitution is specified in terms of other letters from the same alphabet. For instance, given the rules:
- A → ABC
- B → C
- C → D
- D → AC
- E → B
Then a number of expansion steps is specified resulting in patterns being generated from the start symbol 'A'. With the rules above the successive patterns would be:
- A → ABC
- ABC → ABCCDEDEACB
- ABCCDEDEACB → ABCCDEDEACBACBABCDEC
etc. The maximum depth to be set is 8, which would result in a pretty long pattern already with the rules above.

At each positive input step a next symbol from the generated pattern is selected and the corresponding output will be made active. At the end of the pattern it will start again from the beginning. Some rules (or all) can be left blank for simpler patterns. All letters must be from the set [A..E], everything else will be ignored for pattern generation.

This is after an idea found at RobotPlanet, which in turn references this article.

Decided to generate all 8 expansions whenever the rules change, which made it possible to have depth modulation ‐ which I've added.
Made a variation on the previous module generating values instead of events. It also has a couple more rules to compose with. It adds a lookup system so that for each active step a value is put out. The value knobs are mode controlled, one of the modes being 'note value'. There are the usual dB and linear ranges too.

A common chain input is available which is simply added to the current rule based value, and then there are individual rule chain inputs which are added to the output only when the corresponding rule is active ‐ with this the module can be used as a rule based multiplexer too (when all values are set to a neutral C4 or to zero, depending on the selected mode).
$TuneSmithy - A fractal-like note pattern sequencer$ A TuneSmithy module. This is a melody generator after an idea by Robert Inventor and it is explained at the website of the author. The sequence can be arbitrary long in principle, but decided to cap it at 1024 steps. The pattern is a comma separated list of integer values. The second link above explains the inner workings.

in version 2016.8.21.0

When during live morphing new modules are inserted some odd stuff happens. Ok, only when the module does odd stuff, see next item ‐ the GVerb module would crash. Ah, and modules get set to all zero values when inserted while live morph is active. Aha optical mostly … the first variation gets set from the default knobs and all other variations get set to all zeros, which looks confusing in a live morph situation. Still it would be better that on insertion of a new module all the variations would be be set the same ‐ according to the default knob positions. Ok, fixed that.
Live morphing when the GVerb module is present sounds bad (even when it is not being morphed), for FreeVerb it is fine tho. Ah, it was being recreated all the time even when the parameters did not change, and then it would crash. Ok, fixed it. But there also is the issue that automation will change parameters that it should not change. Ok, stopped that too.
Bug in DataGraph where the rightmost point would always end up top right (1,1) even when it would not be drawn there. Fixed.
The active variation randomizer got mixed up a bit with the randomize function from the morpher, unmixed that. The active variation randomizr is selected from the action menu or by pressing 'R' on the keyboard and it sets random values for the current variation only. The randomize function for the morpher sets random values for variations 2 to 6 (for quickly generating a handful of new genes).
Bug in ScaleQuantizer, the selected scale selection is not properly read from a saved patch, the buttons are OK though. Was an issue for all dynamic count selectors and got introduced with the introduction of variations. Fixed now.
Oopsie .. the filter inversions used '-' for 'off' (which can not work anymore after some changes on how dashes are handled for popup menus in relation to automation) changed them to '.'.

in version 2016.8.20.0

Made a Time Of Day (Tod) module which is a saw tooth LFO with a period of 24 hours synced to the local PC clock. So at 00:00 it's output value will be zero and at 01:00 it will be 1 / 24. The clock has millisecond resolution. The output value may drift a bit as windows does time syncs when it feels a need ‐ but it will be the best estimate of the current time. A couple of level shifters and ChebGen modules could be used to obtain signals that run faster. Like first make the signal bipolar, then multiply by 24 (first by 12 then by 2 using two ChebGens in saw mode) and then level shift it again to unipolar positive to get a signal with a rate of one hour. Then the same trick with a factor of 60 to get minutes, 60 again for seconds.

Hmm … instead, I've just added a couple of outs for hour, minute and second.
Made a new option for the morpher, it can do a live morph over the breeder too now. Normal morphing is between the left and the right morph and morphing over the breeder uses all eight variations present in the breeder instead (when the breeder is morphed with the Morph button this is equivalent to the non breeder morph, but when mutate or mate is applied variations will be introduced). This new option tends to (depending on settings and actions) give more lively morph results (like for the XY controls the dot can then move in a rough curve instead of in a straight line only).

And made a new 'soapy' worm shape ‐ how cruel it now seems to hold them in a cage like this …

I've turned the morpher code inside out a bit, that is, I've made variations for patches and based the morpher on that. Variations are stored with the patch, so they will still be there after patch reload. As the variation values are stored in the controls (knobs and such) now, this also means that module additions and deletions do not mess up the worms for other modules anymore. All this needs some work though still; the worms are not always updated correctly, and there is an issue with XY controls loosing data or something ‐ but it stopped crashing at least :-)

Changed the randomizer as well, it now works through the same mechanisms as used by the morpher. It will set random variation values for the active variation now (which then in turn moves the knobs, the randomizer used to work on the knobs directly (which would then change the current variation)). It is all a little easier this way. Anyway … some more work to do on this.

Most stuff is working now, but drag-drop of worms is messy still … some more work …

But it's not working well; turning some more stuff inside out ‐ the mutations are all carried out on the patch now and then update the worms, instead of working on the worms and updating the patch. Let's see what surprises will arise from that …

Ok, that goes better, it seems to all work now as it should ‐ except for it being very slow all. Not sure if its the editor updates or the worm updates … there are a lot of params that change on a live morph … more work :-)

Improved the speed somewhat, still not very fast, but it will do I guess.

Livemorphs are controlled trough the lights update routines, that is not ideal (as the tempo goes all over the place) and this should be changed. Ok, changed that, it is currently updated at a rate of 10 Hz triggered from the audio generation rate (and so independently of the lights update rate). So this should work properly for recording to disk as well. The exact rate may change later from 10 Hz, not sure yet, or make it a user pref maybe.

Live morphing works over the variations now, and I've removed the left/right morph worms. When a left/right morph is wanted you can do a morph operation on the breeder, then copy the worms to the variations and then perform the live morph. So with less options the same functionality is still available.

Added a couple of buttons to the randomizer / morpher screen to make copying easier. Drag / drop now works properly as well. Moved the live morph options from the randomizer to the main toolbar ‐ turn auto mode on or off, and the morph control knob. Morphing no longer changes any variations, instead there now is a temporary variation which is being used.

Changed some keyboard shortcuts related to the randomizer / morpher and added some for variations:
- R - randomizes the patch (was Ctrl+Shift+R).
- Shift + R - toggles the exclusion visibility. (was Ctrl+Shift+Alt+R)
- Alt + R - shows or hides the randomizer window.
(Ctrl+R still is recent patch reload).

For changing variations you can use the keys 1 … 8, or they can be selected by clicking on a worm in the toolbar or from the View menu. Variations can be dragged from one spot to another, or even from or to the randomizer window.

Made it so that it is still possible to see all of the toolbar in the space needed for four horizontal modules … which is like 1024 pixels. This would be a practical minimum value for the horizontal screen resolution needed for Wren (see image above). And also added the mutator commands in there (as well as into the action menu).

Had to change the patch format again … not really for the variations, but could not store the active variation with the patch in the current format. Made the format a little more flexible again, so that future patch parameters will not need a new patch format (which already was the case for control and module parameters). Anyway, it's backwards compatible, but not forwards, as in: old patches can still be read.
Made an Auto module which can perform Auto-randomize, -mute, -mate or -morph in the same way the buttons in the randomizer / Mutator window do. Making this a synth that can truly auto-mate!
The '-' avoidance code has a bug it seems … causing Wren to hang in an endless loop on some patches when the randomizer is used ‐ investigating. Ok, solved that, but still saw dashes on morphing … fixed that too.

Which … of course … introduced an issue for selectors having only two values and no dashes at all, fixed that too.
The PitchShift module can not support randomization on the delay parameter. I disabled that. The problem is that when delay and shift are changed at the same time invalid internal values can occur. All automation has to be and was set disabled for that knob.
When a control was set to be non-randomizable, or non-automatable actually, and in the patch it was stored to allow for randomization the red exclusion rectangle would not show. Fixed that, and when such a patch is saved the control will be saved as non randomizable too.
PitchShift module would produce denormals under some conditions. Probably fixed it with normalizing the linear interpolation code's output.
Gabe made some new bitmaps for the looks. I've made some new ones too – had a lag of about 40. Looks file got updated to 2016.8.2.

in version 2016.7.27.0

Fixed the patch randomizer for selectors having '-' values (these are meant to generate separators in the associated popup menu). Trough the menu interface such items can not be selected, but the randomizer found 'm allright …

Ah, and clicking on a selector or using the cursor keys or the space bar hits them as well. Fixed that too.
Made the Pad, Spiral and DataGraph modules randomizable.
I'm working on a patch mutator, it's very much like what the G2 has. Still some quirks to iron out on that, like what to do when the patch gets modules inserted or deleted. The current implementation does not handle that as it should. Also the images seem to not always be drawn correctly. Ok, found the issue, I'm misusing, sort of, some selectors as indicators (on the ScaleQuantizer). This means that the chord selector has to be disabled for randomization. It should maybe also indicate that the key selection does not map to the selected chord anymore. But then again … I could just consider the selector to be a preset selector only, not reflecting the actual state. On a patch reload the actual keys will win anyway. I may change this when i get ispiration.

I've made a FAQ entry for the morpher / randomizer.

The Wren forum thread may have some info not listed here.

Made the mutator work for Pad, Spiral and DataGraph too. There are some issues with the DataGraph, as points can be added to or removed from it. This means that the number of data points present in the mutator may change. What I did for now is that such a situation is detected and the mutator will be re-initialized (which looses all the current worms in the mutator, making a random set of new ones, and the scratch area will be cleared). Module insertions and deletions are handled in the same way.

Added two extra worms to the mutator, a left and a right morph, and a knob to fade between these two. Also added a modulation option to the knob, and a new module which can do the fade from the patch. This works at the lights rate, which is user controllable, but it can not be very fast nor can the exact timing be very predictable, as a lot of screen updates need to be done. Still for slow morphs between two variations it would be nice. It is possible to have more than one LiveMorph module in the patch, but things may get a bit jumpy with that.

The randomization exclusion works for both the randomizer and the mutator in the same way. There is a keyboard shortcut to visualize the exclusions, and also when the randomuzer / mutator window is active exclusions will be made visible in the patch (with red rectangles around the excluded controls or modules).
Made it so that when patch randomization is requested (as opposed to single control or single module randomization) and there are no modules selected, that all (randomizable) modules are randomized.
Changed some code for the components introducing a couple of interfaces for them: IAutomatable (for MIDI control), IRandomizable (for plain randomization) and ICollectable (for controls that can contribute to the genome used for patch mutation). This makes things a bit easier to program, I hope … it was quite some work. Please note that non automatable controls can not be randomized either. In general anything that needs patch recompilation can not be automated or randomized.

in version 2016.7.23.0

Made a scale control for the Reverb module. It sets the internal length of the delay lines to some factor of the original length. This alters the color and buzziness of the reverb. As the lengths of the internal delays are modified a change of scale needs patch recompilation( e.g. a change causes a brief silence).
Added an End Of Cycle (EOC) output for the AHD and ADSR envelopes. When fed back into the trigger input turns the envelope into an LFO, but can it be used to chain envelopes as well such that one triggers after another.
Moved the Constant, Amplifier and Ratio modules to the Math tab. This distributes the modules a bit more evenly over the tabs.
Working on a patch randomizer. Randomization is possible on Knobs, Sliders and Selectors. It is possible to randomize a single control, all controls on a module, or al controls on a selected set of modules. There will be options to disalow randomization on selected modules and on selected controls. The action menu will get an option to make visible the modules excluded from randomization.

Made a randomizer control wimdow to make it possible to exclude module types and control types, partly by category.
For the randomizer I had to change the patch format to allow for a new module parameter type. Old Wren patches can still be read by the new Wren but new Wren patches can not be read by old Wren versions. I did make the change in such a way that in the future it should be possible to add new module parameter types without altering the patch version. This already was the case for new detail (knob etc.) parameters. Old Wren's should then just skip the parameters they don't know about.

in version 2016.7.4.0

Made a popup menu for the DataGraph, or for it's DataMaker component rather. Some wave shapes can be selected and some basic operations on the data can be selected. With the symmetries there are some duplicates in the operations, like a Bell shape is the same as a vertically flipped cosine, and for the squares there are some duplicates as well. Added load from file and save to file to the menu. Added a left – right adjustment feature as 'make loopable' to the context menu.
MixMS8to1 and MixM16to1 were not working without modulation signals being connected, fixed that.
The timer controlling the lights and VU meter updates is started in running mode. When the startup is slow for some reason this may cause the VU meters to be updated from non‐existing objects. Fixed that by disabling the timer at startup, after each patch compile it was turned on already, so things should be all fine once an actual patch is present.

Decided to make all timers inactive during startup to enable them only in FormShow or even later – when needed. Also typed some extra checks around the VU meter calculator objects to see if they were allocated before being used.

MIDI TX messages were logged from the audio thread directly (when logging was enabled), which was a bug. Fixed that.
Error in DataGraph module. It was meant to react to double left clicks or ctrl+left clicks by adding a new data point, but ctrl+double left click would add two points and would then mess up the order of the points. Fixed that and checked at other places if such a problem could exist elsewhere ‐ did not find such places though.
There is some debug stuff left in the looper module (like the light), this should be removed once the thing gets settled … which currently it is not.
Delayed help file creation a bit so that the looks are completely set up before the images are being created ‐ image previews on the module selector look better now.
Made the trigger LEDs for the envelope modules light up 125 ms always, even for very short trigger pulses. Some more will follow as some modules output very short trigger pulses that could not been seen on all input lights. This also means that for a non changing input signal the light will go off after a while, it will only show the trigger moment. made an error in this making the sample and hold turn into a track and and hold … and fixed that again.
It may be the case that there is a denormals issue for the Formant2 module, running some debug tracking on that, but lost focus … focusing again … After 10 or 11 hours of runtime on a patch, yup, got a denormal report, looking into the why now, it is on the output signal, should not be possible inspecting the code, odd.
Restyled the module documentation a bit and added a link to it in the main menu of this page. Also for newer modules I started to add links from the module images in this page to the documentation page (meaning that module images here are clickable now). Also fixed some inaccurate help text.

I've started using a mark‐down processor for easier markup of the generated help file. See the Acknowledgements section for more on this. This will allow for clickable links in the help, better looking lists, etc., Expect more help restyling.
Changed some things for popup menus. They should now popup near where the cursor is instead of near where the top‐left of the control happens to be. Also in the patch editor I've made the menu key work for the element the mouse cursor is on.
Moved wire toggling and wiggling from the edit to the action menu, keeping the same shortcuts (Ctrl+space bar and Ctrl+W)
Made three new mixer types MixQS2to1, MixQS1to1 and QuadPan (which actually is a panner, but it's on the mix tab …). Although these do have blue connectors they are intended to be used for audio mainly and when audio signals get connected the connectors will change to the audio rate color. The panner uses an equal power distribution only. The mixers allow for 24 dB of amplification.
Made an audio to control rate converter, which can be used to keep large areas of a patch from becoming red.
Added a 'retrig'(ger) option to the LfoTrig module. When set into retriggerable mode and when the trigger input goes inactive and active again during the current pulse train the module will be reset to start the train anew. The default setting is 'normal' for compatibility with older patches. Together with the DataGraph module this could be used to make retriggerable arbitrary envelopes, for instance. Hmm … not quite … anyways …it seems useful … maybe.
Added a Polar module, it translates rectangular coordinates into polar coordinates.
Added a Rectangular module, it translates polar coordinates into rectangular coordinates.
Made a stereo version of the looper. It still has the drawbacks of the mono one, but really wanted stereo.
Added a Spiral module, an experiment really. Per spiral winding of mouse movement a settable amount of note steps is added. Along the positive X axis the white places correspond with the steps. As this internally works with polar coordinates I made outputs for the angle (phi) and distance / radius (norm) as well. The len output represents the spiral length for the current (X,Y) location (from which the note output is calculated). So at least it doubles as a Polar sort of Pad module. (But the latter can also be done by adding a Polar module after a Pad.)
Added an unattenuated frequency control input for the Lfo, SquareLfo and RandSig modules. This makes it easier to add overall speed control to a patch.
Added an MConst module for selecting some more or less often used mathematical constants, like 1, 2, e and 𝜋 etc.
Added a 'wrap' mode to the DataGraph module, the standard behaviour was 'clip', which still is the default. In clipped mode an out of range input value will clip on one of the graph edges and in wrapped mode it will wrap around and come in on the other side again (modulo operation).
Added f(x)=x·𝜋, f(x)=x·2·𝜋, f(x)=x∕𝜋 and f(x)=x∕(2·𝜋) functions to the Monadic function module. Also added f(x)=floor(x), f(x)=round(x), f(x)=ceil(x) and f(x)=abs(x).
The Pad and Spiral module did not set their output values after a module load or patch change, it needed a movement first ‐ I've changed that. I may have thought this to be a feature for the Pad module, but for the Spiral it was just a plain bug.
Removed all the all-too-obvious email addresses from the web page and documentation.
Removed some unused stuff here and there from the code, making the executable a tad smaller.
Added some volume damping around patch recompile operations, the worst of the recompilation click is out now.
I've made an option to disable automatic doc file generation. This option is off by default (so docs will be generated out of the box), and this feature will only work when you use the old wren.ini file, e.g. when you copy the new executable over the old one (or copy the old ini file to the new location). You can always manually generate documentation from the Settings Screen F3 by using the 'generate documentation ...' button there. Note however that when no docs were generated the popups from the module selector will not show a module, or an old version of the image.
I've added command line passing of a patch file name. You can now associate Wren with '.wren' patches in windows such that double clicking on a patch will open it in the Wren instance that got associated. It is still possible to run multiple Wren instances, but they will have to run from different directories now, and patches opened through the explorer will always open in the associated instance. When starting Wren directly it will still open the last patch it had loaded as well. BTW it is possible (it has always been) to copy patches, or parts of it, from one Wren instance to paste them into another) or into a text editor, or whatever program that will accept text.

in version 2016.6.5.0

Changes in text controls and data graphs marked the patch for needing recompilation which was overkill. It should have been marked for change only. Changed that, and internally renamed some 'PatchChanged' stuff into 'Recompile'.
While all modules seem to be properly execution locked now during heap operations the patch as a whole was not locked. Changed that.
MIDI CC assignments to knobs are not working properly ‐ CCs move the knobs alright but the synth value is not changed. Ok fixed that, but now on very fast MIDI I'm getting all kinds of memory issues. Ok, that actually is due to some errors in MIDI transmission ‐ it was doing memory allocations from the audio thread. Ok, four bugs fixed in this … messy, but fixed now.
Added exponential FM control to looper. Oh that it was no longer clicky was a mistaken observation, with long loop times or record being off it is not too noticeable maybe. And a remark about the interpolation mode, all pass mode is not to suitable when the interpolation point moves relatively fast, that is for largish modulations.

in version 2016.5.29.0

The CubicClip module could generate denormal values for small hysteresis and alpha values combined. Fixed that.
When a text editor popup appears the wires of the patch will now be temporarily hidden, and once the editing is done they will be set back to the state they were in when editing started. This solves the problem where text to be edited was not nicely visible when wires happened to run over it.

I've changed the Normalize function to now filter out denormal values, small values, inf's and NaN's. Like:

  function  Normalize(const aValue: TSignal):TSignal;
  type
    TWordArray = array[0..3] of Word;
    PWordArray = ^TWordArray;
  begin
    if
      (((PWordArray(@aValue)^[3] and $7ff0)-$10) >= $7fe0) or
      ( Abs(aValue) < NORMAL_LIMIT)
    then Result := 0
    else Result := aValue;
  end;

This does add a little extra run-time overhead - but less than 1% or so on a 90% patch. The NORMAL_LIMIT value is at -400 dB, TSignal is a Double.

Made text changes mark the patch as being changed, so they can be undone. Also made undo work for the new graph control.
Fixed the patch profiler (only for the profiled versions, not for the normal release version) to work properly for RateSmart modules. (Profiling adds about 8% performance overhead, so it is turned off for release versions.)

And then removed a lot of needless profiling info from the report.
The 1 kHz fader for the third filter bank was wrongly named and hence that band was not working at all - fixed it.
The GVerb module was not properly locked against execution when it was being recreated, fixed that one and two or three other cases as well. Looks like overall stability got improved with that, in that undo/redo keeps working etc.
Error in MIDI running status processing. Only the first two bytes for messages based on running status would get interpreted. As for $90 $01 $02 $03 $04 only a note‐on with velocity 2 for note 1 would be produced and the note 3 with velocity 4 note‐on event would never get processed. Fixed that.
Sped up patch execution by about 5% trough removing some dynamic type checks from the inner execution loop. And by another 2% or so by skipping some profiling related code for the release version of the code. Sped up the dezipper code a bit too ‐ but that's not really noticeable (as to be expected I guess as it works at control rate).
Added filter inversion for all filter bank modules. The idea is to invert every odd filter before the summation occurs and to not do that for the even filters. After doing some filter sweeps this does not seem to alter the filter response much though … it gets a little less bumpy (judged by listening), but then again it does not cost much. processor wise, so will leave it in anyway.
Made a clock generator module that acts like the Clavia ClkGen module. It can not sync to MIDI for now. The blue outputs have the internal phase accumulator values, these can for instance be used to implement your own swing functions by feeding them into a comparator.

Added an input for external tempo control, this overrides the knob setting. This makes it possible to control the tempo with the Tapper module, which got a new BPM ClockGen mode added for that purpose.
Made a count module, it counts upgoing signal edges on it's count input and it displays the total count. A frequency measurement tool would be handy as well (which would be a counter with automatic periodic reset).
OK, made a frequency counter module as well, it counts upgoing signal edges on it's count input and it displays the total count over the measurement period recalculated to Hz. It is a bit of a primitive thing ‐ like you have to wait for 10 s to get 0.1 Hz resolution. But seems handy to have still, and maybe will expand a bit on it at some later stage.
Yet Another Random Sequencer! When quantize is on the internal values will be quantized according to the note fraction quantization value selected. When quantize is off the plain random values will be used as the internal value. The chain input is never quantized but just added to the current internal value after all other processing. The built in slew rate limiter can be applied either pre- or post-quantization, it does not act on the chain‐in value.
Experimenting with speech synthesis a bit. Made a very simple speech module based on the Tiny Speech Synth. In the current setup it translates a sequence of phonemes to an internal wave file that can then be played just like the wave player does.

Will need to rework this a bit to make it create sound on the fly. The idea is to make something like shown in the image (the 'song' module below). The module should flag text start and phoneme starts. With these signals sequencers can be driven to control the speech rate, the oscillator pitch and the amount of whisper.

Getting there slowly … re-implemented the internal oscillator, so it can be properly tuned. Still struggling a bit with extreme duration modulations that are causing crackle ‐ phoneme overlap is a bit of a PITA. And speech is not very intelligible, but not worse than say a human voice trough a G2 vocoder … maybe.

I've changed the 'duration' control into 'speed', added oscillator wave form selection (saw or pulse, both anti‐aliased). Added a filter configuration control as there was some odd pole or zero stuff going on in the normal mode which still may be interesting … but it can be turned on or off now. This works by inverting every odd filter's output before summation in the 'FI' mode.

Changed the speed modulation a bit, but it is still not perfect ‐ clicky at times. The PM I've added is quite useful though. Also I've added some more outputs, one that goes active on each 'silence' and another going active on each 'vowel'.

I've made the default speed quite low (or actually the phonemes are quite long), to allow for very slow textures, but it can be sped up by a factor of 10 or so when using external speed modulation.
Made a VOSIM thingie. It is meant to be driven by an oscillator to set the rate and it can then control a single formant. Added 'Whisper' to make noisy tones, and AM to set the volume dynamically. The number of pulses can be modulated as well from one to the set number. The decay parameter controls the attenuation after each output squared sine pulse. The number of pulses and the decay factor determine the formant strength. The formant frequency is controlled with the formant parameters.
Working on a new type of graphic control which can be used as a lookup element. For example in this new DataGraph module. The input would expect a [0,1], [-1,0] or [-1,1] signal and the output would then output the Y value for that X value also [0,1], [-1,0] or [-1,1]. It should be possible, when finished, to add points to and remove points from the graph. It would probably do linear interpolation to find Y values, but it's just an idea for now – working on it … currently it's an image.

Ok, it is a working control element now that can be used on modules. Points can be added by double (or control) clicking into an empty area. Double or control clicking on a point allows the point to be moved. Removal of a point is just a move to somewhere out of the control. There is a minimum of two points in the control, one on the very left and one on the very right – these can only be moved up or down and can not be removed. There is a restriction on horizontal movements too, a point can not be moved to after or before an existing other point.

When moving a point a crosshair cursor is shown, and in play mode a vertical bar cursor will be visible to indicate the current playback position. Not sure yet how that will work out at audio rate playback. Now to get a first working module with it … not so bad … good actually, the moving vertical bar works, now the rest.

The compiled synth now gets updates from the control on patch load and on point moves, just a little more work now in updating the synth data and making an interpolator for it.

We have lift-off! After fixing some memory issues, of course, and some small tweaks … like decorating the dots a little and making it a transparent control :-)

And now with splines too. Turns out to be not too bad for lower audio rates :-)
Made a type flipper with slew control. A fixed input signal type can be set and for the output a left and a right type can be selected. When the select input is low the left output type will be applied to the left output and the right type to the right one. This is flipped when the select input goes high. The speed of type flipping is controlled with the slew parameter ‐ when slew is set low the switch is immediate, when it is set to the maximum value the flip never actually happens. The selectable signal types are: 'normal', 'inverted', 'positive', 'positive-inverted', 'negative' and 'negative-inverted'.
Working on a looper thingie - it is a mess right now … well it basically works but it's clicky still. Added some mixing around the end loop to begin loop transition regardless of whether the playback pointer moves forwards or backwards. This helps somewhat … The amount of overlap can be set with the overlap control from zero to one.

Changing the pointers to be relative to the in‐pointer helps more. Sounds like it is click free now. Loop readout interpolation can be set to either linear or all pass mode. Both are having different artifacts, but linear seems to work better for long delay times and all better for short times. The maximum time can be set from 110 µs to 5 minutes. Duration and start controls are set as a fraction of the maximum time. Speed can be set from -4 to 4, negative speeds meaning the playback is backwards. For longer loop durations the modulation of start and duration are quite sensitive, decided to make the amount controls to work logarithmic.