ENGINE AWavelet Subtractive

1The idea in one minute

Partials versus atoms

A classic synthesizer thinks in partials. A sawtooth is a stack of sine-wave harmonics at 1, 2, 3, 4 … times the note frequency, each quieter than the one before, and a low-pass filter turns down the harmonics above its cutoff. Every harmonic lasts for the whole cycle, so this description says how much of each frequency there is — but nothing about where in the waveform it happens. The saw’s sharp reset and its smooth ramp are mixed into the same partials.

A wavelet transform describes the same cycle with atoms: short wiggles that all have the same shape — the wavelet — and come in sizes an octave apart. The largest atom spans the whole cycle; the next size has two atoms, each half a cycle long; then four, eight … down to 1024 tiny atoms spaced two samples apart. All atoms of one size form a band. Because each atom sits at a definite place in the cycle, the transform tells you both which octave and when.

• One cycle = 2048 samples = 2047 atoms (1 + 2 + 4 + … + 1024) plus the average, which is removed.
• Band b has 2^b−1 atoms. Band 1 is about the fundamental, band 2 the 2nd harmonic, band 3 harmonics 3–4, band 4 harmonics 5–8 … each band is about one octave of harmonics.
• Smooth stretches of a wave need almost no fine atoms; sharp edges need many. That is why only the reset edge of the saw lights up the fine rows in Figure 1 — and why the POSITION window can treat edges separately.

Engine A analyses your source wave once, splits it into 11 band layers — the part of the wave carried by each band — and plays them back together, each multiplied by its own gain. Setting those gains is the job of the SCALE FILTER.

Where the bands are

The cycle is analysed once and played back at the pitch of each note (it is pitch-synchronous), so the bands are relative to the note: band 4 is always about harmonics 5–8, whatever key you press. Key tracking is built in. In hertz, the bands move with the note:

“—” = not played at that pitch (see below). Values are for the smooth wavelets; Haar and the short Daubechies bands overlap more with their neighbours.

To keep high notes free of aliasing, each band layer is stored in eleven band-limited versions, one per octave, and every note uses the version whose harmonics stay below the sample rate minus 19 kHz (about 25 kHz at 44.1 kHz, 29 kHz at 48 kHz) — anything that folds back lands above 19 kHz. The consequence is musical: the finest bands only exist on low notes. Every octave down adds one band. At 48 kHz, band 7 is the highest band played for notes from about A♯3 to A4, band 8 an octave lower, band 9 two octaves lower; band 10 needs notes below about 57 Hz and band 11 below 28 Hz. (At 44.1 kHz the limits are about a whole tone lower.)

The wavelet chooses the atom

The WAVELET selector (shared by both engines) decides what the atoms look like. Seven classic wavelets are available, from the 2-tap Haar step to the 40-tap Daubechies 20:

The trade-off is the classic one between time and frequency. Short wavelets are precise in time — edges stay sharp and the POSITION window has crisp borders — but their bands overlap, so each band also contains some neighbouring octaves. Long wavelets separate the octaves cleanly but each atom is spread over more of the cycle. For subtractive sweeps that sound like a conventional filter, choose Daubechies 8 or longer; for stepped, digital colours choose Haar or Daubechies 2.

Signal flow

• Table building happens only when you change WAVE, WIDTH, PHASE, SPARSITY or WAVELET: the source cycle is drawn (moved by SOURCE PHASE), transformed, thinned out by SPARSITY, and split into 11 band layers plus a “dry” layer (their sum, i.e. the unfiltered wave). The switch to new tables is crossfaded, so you can change these controls while playing.
• Every voice plays these layers at its own pitch and mixes them with the SCALE FILTER gains, which are recalculated continuously from CUTOFF, the envelopes, the LFOs and velocity. The POSITION window then decides, point by point within the cycle, how much of the filtered sound replaces the dry one.
• All voices are summed and pass through the effects — saturation, chorus, delay and reverb, in that order — a DC blocker, the master volume and a soft limiter that keeps the output below 0 dBFS.

A BEAUTIFUL CONSEQUENCE

The square wave that will not filter

With the Haar wavelet, the band-1 atom is one cycle of a square wave: high for the first half, low for the second. So when you analyse a SQUARE with Haar, the whole cycle is described by exactly one atom — the other 2046 are zero. The SCALOGRAM shows it: only the bottom row lights up.

The low-pass filter therefore has nothing to remove. CUTOFF does nothing until it reaches band 1; below that the sound simply gets quieter. For progressive subtractive sweeps on a square, choose a smooth wavelet — Daubechies 8, Symlet 8 or Daubechies 20: their atoms do not match the square, so its edges spread over all bands and the filter has something to work on. Or use the effect on purpose: Haar + SQUARE is a tone no filter can dull. The same happens, almost, with PULSE at 25 % or 75 %: the edges fall on quarter-cycle boundaries and Haar needs only two atoms.

2Quick start

From closed to sounding in five steps — pick the column for the version you use.

• Run the demo plays the default demo — Beethoven’s Appassionata as a tour of the presets, about two minutes — exactly as ▶ DEMO in the header does (see Demo).
• Play the synth (“MIDI supported”) only starts the audio: no note is played. Then play with the mouse, the computer keyboard or a MIDI keyboard.
• Download the VST opens the release page of the free plug-in (Windows VST3 / VST2, macOS VST3) in a new tab; the welcome window stays open.
• A click outside the window also starts the audio. The window comes back, with a message in place of its sub-title, when another window or tab of the app takes the audio (see One window at a time).

3Panel tour

The panel is the same in the web app and the plugin, apart from a few buttons: the web app’s ▶ DEMO and DOWNLOAD VST, the plugin’s window-size button, SAVE PRESET and the SYNC buttons of the tempo sync. From top to bottom (beta 0.6): the header; the engine section, SUB OSC and SCOPE; SCALE FILTER, POSITION and the two envelopes; LFO 1, LFO 2 and EFFECTS; VOICE, the wheels and the keyboard (beta 0.7: a taller bottom row, with the credit line “By OktoPierre · beta 0.7 · GPL v3” under VOICE). In engine A mode the accent colour is teal and the upper-left section shows the ENGINE A controls. The screenshot shows the factory preset A · Mirror Sub Bass Shift (a sound of the Source Phase section); click a number to jump to its description.

1234567891011121314151617181920212223242526272829

1. 1▶ DEMO ▾ (web app) — ▶ DEMO plays the default demo, a tour of the presets; ▾ opens the list of demo pieces.
2. 2ENGINE — chooses the engine panel shown, A (teal) or B (amber); A/B MIX sets what you hear.
3. 3Preset browser — ◀ ▶ step through the presets; click the name for the list.
4. 4DOWNLOAD VST (web app) — opens the release page of the free plug-in; the plugin has SAVE PRESET here.
5. 5MANUAL — opens the manual of the engine shown: this one, or Manual B on the B tab.
6. 6VOLUME and meter — master level and stereo peak meter.
7. 7WAVE — the source waveform of engine A.
8. 8ATOM SHAPE — one atom of the selected wavelet.
9. 9WAVELET — chooses the atom shape (seven wavelets); WAVELET 2 below it, for the WAVELET MORPH.
10. 10SOURCE CYCLE — the wave before (grey) and after sparsity, with “K / nz ATOMS KEPT”.
11. 11WIDTH — pulse width or triangle symmetry; greyed out unless WAVE is PULSE or TRI.
12. 12SPARSITY — keeps only the largest atoms.
13. 13PHASE — SOURCE PHASE: moves the wave inside the cycle before the analysis (beta 0.4).
14. 14A/B MIX — what you hear: engine A, engine B or both mixed.
15. 15SUB OSC — LEVEL, OCT and SHAPE of the sub oscillator, one above the other (beta 0.7).
16. 16SCOPE — one cycle of the output; grey = the dry wave.
17. 17SCALOGRAM — the atoms of the 11 bands; dashed line = window.
18. 18SCALE FILTER — per-band filter with resonance, key tracking, envelope and velocity.
19. 19POSITION — confines the filter to part of the cycle.
20. 20AMP ENVELOPE — loudness ADSR and velocity sensitivity.
21. 21MOD ENVELOPE — ADSR that moves the filter cutoff.
22. 22LFO 1 — shape, RATE and how far it moves the pitch, cutoff, position and wavelet morph; SYNC in the plugin.
23. 23LFO 2 — a second LFO on the parameter chosen in its TARGET menu; SYNC in the plugin.
24. 24EFFECTS — saturation, chorus, ping-pong delay and reverb, in the order of the sound; the delay’s SYNC in the plugin.
25. 25CHORUS — string-machine triple chorus: amount, RATE and DEPTH.
26. 26VOICE — portamento mode and voice LEDs in its title row; octave, tuning, polyphony, glide and unison below.
27. 27Wheels — pitch bend and modulation (vibrato).
28. 28Keyboard — five octaves; click lower on a key to play louder.
29. 29Status bar (web app) — audio and MIDI status, key help, Save preset / Export / Import.

The header

• ▶ DEMO (web app only) plays the default demo, a piece during which the factory presets change every few seconds; the ▾ beside it opens the other demo pieces and lets you load your own MIDI file — see Demo.
• A · SUBTRACTIVE / B · ADDITIVE chooses which engine section is shown in the upper left (and the accent colour). What you hear is set by the MIX knob at the right of that section — engine A, engine B, or both mixed (see Mixing engines A and B). The sub oscillator, filter, position, envelopes, LFOs, voice and effects are shared by both engines.
• The preset browser shows the current preset; a dot after the name means it has been modified. See Browsing and saving presets.
• DOWNLOAD VST (web app only, beta 0.6) opens the release page of the free plug-in — Windows VST3 / VST2 and macOS VST3 — in a new tab. In the plugin, SAVE PRESET sits there (see the plugin’s differences).
• MANUAL (beta 0.6: one button instead of MANUAL A and MANUAL B) opens the manual of the engine shown — this manual on the A · SUBTRACTIVE tab, Manual B on B · ADDITIVE; each manual links to the other. In the web app it opens as an overlay (close it with ✕, Esc or a click outside; “Open in new tab” pops it out), in the plugin in your web browser.
• The small knob at the right is the master VOLUME (−36 to +6 dB, default −3 dB), followed by the stereo output meter. In the plugin, a size button next to the logo opens a menu of window sizes (70–150 %).

Working with the controls

4Engine A: source and sparsity

The ENGINE A section creates the raw material: a source wave, its analysis into atoms, and an optional thinning-out of those atoms.

Source waves

Sparsity: a lossy codec as a timbre control

SPARSITY ranks the atoms of the cycle by size and keeps only the K largest; the rest are set to zero. This is exactly what wavelet-based lossy codecs such as JPEG 2000 do to save bits — here it is a sound-design tool. At moderate settings the cycle simplifies into a handful of atoms and their shape becomes audible: glassy, slightly “digital” tones; at 100 % only one atom is left.

The range is relative to the source. A smooth wave needs far fewer than the 2047 possible atoms — a saw analysed with Daubechies 8 has only 96 non-zero atoms, because a straight ramp needs no fine atoms — only the jump does. SPARSITY counts from that number, nz: K = nz^{1 − SPARSITY}, rounded. So 0 % keeps every atom the wave needs, 50 % keeps about √nz, 100 % keeps one, and the knob does something over its whole range for every wave.

Atoms kept for a few sources, as “K / nz” exactly as the SOURCE CYCLE display shows them.

• Coarse or fine steps. Sources that need few atoms change in clear jumps — a Daubechies 2 SAW has only 20, and Haar + SQUARE has a single atom, so there is nothing to remove at all. NOISE, and any slope analysed with Haar, use all 2047 atoms and change almost continuously.
• SPARSITY acts before the filter, and the dry signal used by the POSITION window is the sparse version too.
• Each change rebuilds the tables (crossfaded), so turning the knob while playing gives a stepped morph rather than a smooth sweep.

Source phase: moving the wave inside the cycle

PHASE (beta 0.4) slides the source wave along its cycle before the wavelet transform: at 90° the saw’s jump sits a quarter of the way in instead of at the start. To the ear an unfiltered wave sounds exactly the same at any phase. But the atoms of the wavelet transform sit on a fixed grid inside the cycle, so the moved wave is described by other atoms — and as soon as something acts on the atoms (a SCALE FILTER cut, SPARSITY, the POSITION window), the result is a different timbre. A Fourier filter, which only sees harmonics, cannot tell the two phases apart; a wavelet filter can. This is the shift-variance of the discrete wavelet transform, turned into a control.

• Which wavelet. Short wavelets react most: Haar and Daubechies 2/4 give changes of tens of dB on some harmonics; Daubechies 20 only a few dB. The table below measures it.
• Which bands move. Band b has 2^b−1 atoms, one every 2^12−b samples. A shift by a multiple of that spacing slides the wave exactly onto the next atoms of band b, so that band is unchanged; only the bands with a wider spacing change. 180° (1024 samples) leaves bands 2–11 as they were; a small odd shift (say 37 samples, 6.5°) changes them all.
• When it does nothing. With the filter wide open, no sparsity and no position window the sound is the same at every phase (the layers always add up to the moved wave). Make the filter cut into the strong low bands (CUTOFF 2–5, a steep SLOPE, RESO), add SPARSITY or the window, and turn PHASE.
• Noise is moved by whole samples. The SOURCE CYCLE display shows the moved wave, so you can see where the edge now sits.
• Moving it with LFO 2 (TARGET PHASE): AMOUNT moves the phase by up to ±180°. The tables are then also built at 11 more phases, 30° apart round the cycle, and each voice crossfades between the two nearest — a smooth scan through the timbres of the cycle (preset “A · Phase Scanner”). This costs about 12 times the table memory (about 13 MB, twice that with WAVELET MORPH) and up to twice the processing per voice; at AMOUNT 0 nothing extra is built.

Largest change of a harmonic (1–200) against 0°, computed from the engine. 1024 samples is a multiple of band 6’s atom spacing (32 samples): band 6 slides onto its own atoms and nothing changes.

The presets of the Source Phase section (beta 0.7: the last of the preset browser again; it was second, after TOP Sounds, in beta 0.5 and 0.6) are built on this control: phased versions of OktoPierre’s favourite and best-voted sounds (“… Shift”) and new sounds. Of the 100 of beta 0.4 he kept 44; since beta 0.7 his TOP Sounds are only his own selection, without Source Phase sounds.

5Scale filter

The SCALE FILTER is the heart of engine A. It does not work on frequencies directly: it gives each of the 11 bands its own gain, and the sound is the sum of the band layers, each multiplied by its gain. A low-pass therefore removes the fine atoms — the small, sharp details of the cycle — and keeps the coarse ones. Because one band is about one octave of harmonics, the controls behave like those of a familiar synth filter: CUTOFF in bands, SLOPE in dB per band (≈ dB per octave), resonance, key tracking and envelope.

Two things are different from an analogue filter. There is no phase shift and no feedback, so a filtered edge becomes wider and smoother but stays exactly where it was in the cycle, and the resonance can never self-oscillate. And the character of a sweep depends on the WAVELET: with Haar every removed band halves the number of steps in the wave, with Daubechies 20 the bands are nearly pure octaves and the sweep sounds smooth.

LFO 1 CUTOFF adds a further ± bands of periodic movement, and LFO 2 can move CUTOFF or another filter control — see LFO 1 and LFO 2.

The four types in numbers

• Low-pass — bands below the cutoff pass untouched; the band at the cutoff is about 0.17 × SLOPE down (−4 dB at 24 dB/band); each band further up loses about SLOPE dB more. At CUTOFF 12 (the default) every band is within 0.1 dB at 24 dB/band (0.4 dB at 96 dB/band): fully open.
• High-pass — the mirror image: bands above the cutoff pass, bands below lose about SLOPE dB per band.
• Band-pass — keeps a window about one band wide: at 24 dB/band the cutoff band is −0.8 dB, its neighbours −13 dB, two bands away −36 dB.
• Band-reject — removes the band at the cutoff completely. The notch width depends only on SLOPE: at 6 dB/band it hollows out two or three bands on each side, at 24 its neighbours are only 1.6 dB down, at 96 nothing but the exact band is touched. With steep slopes, a cutoff between two bands (say 5.5) hardly notches anything — use whole numbers.
• Resonance — adds up to +24 dB centred on the cutoff, falling off over about ±1 band. With a low-pass at 24 dB/band and RESO 12 dB, the cutoff band ends up at about +8 dB.

One special case deserves a warning: with the Haar wavelet a SQUARE source is a single atom, so the filter has nothing to remove — see The square wave that will not filter.

6Position window

A classic filter treats every part of the waveform the same way. Wavelet atoms, however, live at definite places in the cycle — so engine A can filter one part of the cycle and leave the rest untouched. This is something an ordinary filter cannot do: soften only the falling edge of a square, darken only the reset of a sawtooth, or sweep a “filtered zone” around the cycle with LFO 1.

Where the edges are

Most of the action is at the edges of the source wave, where the fine atoms are. They sit at fixed positions:

High notes and modulation

A narrow window multiplies the sound by a short bump, which adds high harmonics. To keep this alias-free, the window is automatically widened on high notes: it is never narrower than 16 × note frequency ÷ sample rate of the cycle — about 9 % at C4 and 35 % at C6 (48 kHz). The minimum WIDTH of 5 % is only reached below about 150 Hz.

LFO 1 POSITION moves the window up to ±180° — all the way round the cycle — and is only audible when DEPTH is above 0 %. The SCOPE shades the window and the SCALOGRAM draws it as a dashed line.

7Envelopes

Two ADSR envelopes run for every voice. The AMP ENVELOPE shapes the loudness; the MOD ENVELOPE moves the filter cutoff, by the amount set with ENV AMT in the SCALE FILTER (it has no other destination and ignores velocity — use VEL AMT for velocity-dependent brightness).

AMP ENVELOPE

MOD ENVELOPE

• Attacks are linear; decays and releases are exponential, which sounds natural and avoids clicks. The small curve above the knobs shows the shape on a logarithmic time axis, so short and long settings are both readable.
• Playing a note on a voice that is still sounding restarts the attack from the current level, without a click.
• With VOICES at 1 (mono, legato), notes played while another key is held do not restart the envelopes.

8LFO 1 and LFO 2

Two LFOs sit side by side in the third row of the panel (beta 0.6): LFO 1 at the left, with its shape tabs and five knobs — RATE, then how far it moves the PITCH, the CUTOFF, the POSITION window and the WAVELET morph; LFO 2 at its right, the same size, moves one parameter chosen in its TARGET menu. Both are shared by all voices. They run freely and are not restarted by new notes, so in a chord every voice moves together — ideal for pads and vibrato. For a movement that starts with each note, use the MOD ENVELOPE instead.

LFO 1

The mod wheel (on screen or MIDI CC 1) always acts through LFO 1: it adds up to 50 cents of vibrato depth on top of PITCH, using the current SHAPE and RATE. With PITCH at 0, the wheel alone brings in vibrato.

LFO 2 (beta 0.3)

A second free-running LFO, in the card at the right of LFO 1, moves one parameter of the SCALE FILTER or of the POSITION window (or engine A’s PHASE) — the one chosen in its TARGET menu. Like LFO 1 it is shared by all voices. beta 0.6: the card has the size of LFO 1’s, with the same shape tabs.

Tempo sync (plug-in only, beta 0.6)

In the plug-in (VST3 and VST2), a SYNC button sits at the right of the LFO 1 and LFO 2 titles. When it is lit, that LFO’s RATE knob becomes a note value at the host’s tempo — Ableton Live’s or your DAW’s; 120 BPM when the host gives none — and shows the value as its label, for example “1/8D”. One LFO cycle then lasts that note value, and the speed follows the tempo changes of the song. The delay has the same button: see Delay tempo sync.

• Note values, from slow to fast: 4/1 2/1 1/1 1/2D 1/2 1/4D 1/2T 1/4 1/8D 1/4T 1/8 1/16D 1/8T 1/16 1/16T 1/32. 4/1 = four whole notes (four bars of 4/4) per cycle; T = triplet (2/3 of the note), D = dotted (1.5 × the note).
• Speed, not phase. SYNC sets the speed only: the LFO keeps running freely and is not restarted on the beat or the bar.
• Off by default. SYNC is off in every factory preset, so older presets, plug-in states and Live Sets sound the same as before.
• Web app. A browser page gets no tempo: the web app has no SYNC buttons and its RATE stays in hertz.

9Voice

The VOICE section sits at the bottom left, before the wheels and the keyboard (beta 0.6): the portamento selector and the voice LEDs in its title row, its seven knobs below. The small credit line under it reads “By OktoPierre · beta 0.7 · GPL v3” (beta 0.7).

• Voice allocation. When all voices are busy, a new note takes over the oldest released voice first, then the oldest held one. Playing a note that is already sounding re-uses its voice.
• Mono and legato. With VOICES at 1, holding a note and playing another changes the pitch without restarting the envelopes (it glides if GLIDE is above zero and the portamento selector is not OFF). Release the new note while still holding the first and the pitch returns to the first one. Legato means that another key is still physically held: with the sustain pedal down but no key held, a new note restarts the envelopes.
• Polyphonic portamento (POLY). With GLIDE above 0, every voice glides from its own previous note. Playing one note at a time, the new note glides from the previous one (the voice of the last note is reused) while the release tails of older notes keep sounding. Playing chords, each new note takes the free voice whose last pitch is nearest, so chord changes glide voice by voice — from C–E–G to F–A–C, G slides to F, E to A, and C stays on C. A voice that has to be stolen glides from wherever it currently is.
• Level. Each note plays at about −8 dB to leave headroom for chords, and unison copies are scaled by 1/√n so that stacking keeps the loudness roughly constant.
• Voice LEDs. The 16 dots show the voices: dim = available, lit = sounding (including release tails). They sit to the right of the portamento selector, in the title row of VOICE.

10Sub oscillator

The SUB OSC section adds a simple oscillator one or two octaves below every note — the classic way to give a sound more low end, from a warm bass foundation to infra-bass you feel rather than hear. It sits in the top row, between the engine section and the SCOPE (beta 0.6; it was at the bottom before): its three knobs one above the other, LEVEL, OCT and SHAPE (beta 0.7; LEVEL was a large knob before).

• One sub per voice. Every note gets one sub oscillator, centred in the stereo field — not one per unison copy, so UNISON, DETUNE and STEREO neither thicken nor spread it.
• Outside the wavelet engine. The sub is added after the SCALE FILTER and the POSITION window, so none of them affects it: it stays clean and steady while you sweep the wavelet sound. Then the AMP ENVELOPE (and velocity) shapes it like the rest of the note, and it goes through the effects — DRIVE saturates it too.
• How low? On C2 (65 Hz) the sub sits at 33 Hz (−1 octave) or 16 Hz (−2 octaves); with OCTAVE at −1 it goes one more octave down. Below about 20 Hz you feel rather than hear it, and laptop speakers, small monitors and most headphones cannot reproduce it at all. The output’s 10 Hz DC blocker softens the very lowest frequencies (about −1.5 dB at 16 Hz, −4 dB at 8 Hz).
• The SCOPE and SCALOGRAM show the wavelet sound only; the sub is not drawn.

11Mixing engines A and B

Since beta 0.2 the two engines can play together. The MIX knob sits at the right of both engine sections, so it is always in reach.

• The tabs choose the panel, MIX chooses the sound. A · SUBTRACTIVE and B · ADDITIVE only choose which engine section is shown, to edit it; switching them neither changes the sound nor marks the preset as modified. Turn MIX towards B, then click B · ADDITIVE to shape the additive layer.
• One note, two layers. Each note plays engine A’s filtered source and engine B’s atoms at the same pitch, with the same AMP ENVELOPE, glide, unison and effects. Both layers go through the SCALE FILTER (engine B also keeps its drawbars, TILT and SCALE ENVELOPE) and through the POSITION window. The sub oscillator is added once, whatever the mix.
• Linear crossfade. At “A 50 · B 50” each layer plays at half level, so the mix stays about as loud as either engine alone. Turning MIX while notes sound is smoothed (20 ms) and does not click.
• Cost. In between, every voice is computed twice, so the CPU load roughly doubles. The tables of the second engine are built when MIX leaves A or B (and dropped when it returns there), with the same 10 ms crossfade as the other table changes.
• Older sounds. Presets, user presets and Live Sets made before beta 0.2 have no MIX: they get MIX = their engine and sound exactly as before. The factory presets that mix both engines are named “AB · …”.
• The SCOPE and SCALOGRAM show the engine whose panel is open.

12Effects and master

The EFFECTS section, at the right of the third row, shows the effects in the order of the sound, each under its heading (beta 0.6): SATURATION (DRIVE) · CHORUS (CHORUS, RATE, DEPTH — they had their own card before) · PING-PONG DELAY (TIME, FEEDBACK, DLY MIX) · REVERB (SIZE, REV MIX). They process the sum of all voices — wavelet sound and sub oscillator together — in this order: saturation → chorus → ping-pong delay → reverb → DC blocker → master volume → soft limiter. The delay and reverb mix knobs add the effect on top of the unchanged dry sound (there is no dry/wet crossfade), so raising them also makes the sound a little louder. The reverb receives the echoes of the delay as well.

• Ping-pong delay. The mono sum of the voices enters the left channel; each repeat crosses to the other side and passes a gentle low-pass, so echoes bounce between the speakers and darken as they fade.
• Reverb. A Freeverb-style stereo reverb (eight comb and four all-pass filters per channel) with fixed high-frequency damping; SIZE sets the decay.
• Output stage. A 10 Hz DC blocker removes any offset, then the master VOLUME is applied, then a soft limiter: it is transparent below −3 dBFS and never lets the output exceed 0 dBFS, so the instrument cannot clip your audio interface or your Live track.
• Meter. The stereo meter shows the output peak over the top 48 dB; the top of the bars is 0 dBFS. If the bars sit near the top, the limiter is working: lower VOLUME, RESO or the number of unison copies for a cleaner sound.

Chorus (string machine)

The chorus (beta 0.2) is a triple chorus modelled on the ensemble effect of 1970s string machines. Its knobs — CHORUS (the amount), RATE and DEPTH — are the second group of the EFFECTS section, under the CHORUS heading (beta 0.6; they had their own card at the bottom before).

• How it works. The sound (after DRIVE) feeds three short delay lines. Each is swept by the same slow and fast LFOs, shifted by 120° from one line to the next, so the three copies are never in step. They are spread left, centre and right: the result is wide and lush even from a single voice.
• Where it sits. After the voices and the DRIVE, before the ping-pong DELAY and the REVERB — the order of a classic string machine going into an echo and a hall.
• Mono in, stereo out. The chorus works on the mono sum of the sound; at 100 % the stereo image is made by the three lines. Lower settings keep part of the original stereo spread of UNISON and STEREO.
• Try it on “A · String Machine”, “A · Solina Cellos”, “B · Ensemble Organ” or “AB · Analog Strings”.

Delay tempo sync (plug-in only, beta 0.6)

In the plug-in (VST3 and VST2) a SYNC button sits at the right of the PING-PONG DELAY heading. When it is lit, the delay’s TIME knob becomes a note value at the host’s tempo — Ableton Live’s or your DAW’s; 120 BPM when the host gives none — and shows the value as its label, for example “1/8D”. The echoes then follow the tempo changes of the song. The two LFOs have the same button: see Tempo sync.

• Note values, from short to long: 1/32 1/16T 1/16 1/8T 1/16D 1/8 1/4T 1/8D 1/4 1/2T 1/4D 1/2 1/2D 1/1. T = triplet (2/3 of the note), D = dotted (1.5 × the note). At 120 BPM, 1/8 = 250 ms and 1/1 = 2 s.
• Up to 4 seconds. The delay holds up to 4 s (a whole note at 60 BPM); a longer note value, at a slow tempo, is held at 4 s.
• Off by default. SYNC is off in every factory preset, so older presets, plug-in states and Live Sets sound the same as before.
• Web app. A browser page gets no tempo: the web app has no SYNC button and its TIME stays in seconds.

13Displays

The SCOPE section fills the top right of the panel, at the right of SUB OSC: the scope on top, the scalogram under its SCALOGRAM label (beta 0.6: the section’s title is just SCOPE; beta 0.7: the section is wider and the scope taller).

The SCOPE and SCALOGRAM draw the cycle as it is stored, before the per-note band limiting described in Where the bands are; on high notes the finest rows are not actually played. The sub oscillator is not drawn.

14Sound-design recipes

Each recipe starts from a factory preset and uses only the controls on the panel. Values are what the panel shows when you hover over a knob.

15Factory presets

108 factory presets show what engine A can do, from clean sounds to extreme effects. Load one, play, then look at the panel to see how it is made — the table lists every setting that differs from the defaults. In the instrument’s preset menu OktoPierre’s favourite sounds come first, then all the presets by category (bass, lead, keys, organ, pad, strings, brass & wind, pluck, bells & mallets, FX, drums & percussion, the Init presets, then — last again since beta 0.7 — the Source Phase section of beta 0.4). This table lists, category by category, the presets that open on engine A’s panel (★ = one of OktoPierre’s favourites), including the “AB · …” presets that mix both engines (their settings show both engines). The drum and percussion presets are pitched: play them on different keys. The Effects column rates how much each preset relies on effects and movement (delay, reverb, saturation, unison, LFO, sub oscillator): none light medium heavy extreme.

Browsing and saving presets

The preset browser sits in the header. The 150 factory presets start with OktoPierre’s 22 TOP Sounds (beta 0.7: exactly his selection, from B · Mirror Sub Bass to A · Laser Sweep, without Source Phase sounds); the others follow sorted by category, like the banks of a hardware synthesizer — Bass, Lead, Keys, Organ, Pad, Strings, Brass & Wind, Pluck, Bells & Mallets, FX, Drums & Perc, Init and Source Phase — so the two Init presets come just before the 44 Source Phase sounds, which are last (beta 0.7; they were second in beta 0.5 and 0.6). ◀ ▶ step through them in that order (and then, in the web app, through your own). Clicking the name opens the list: in the web app ★ TOP SOUNDS at the top, then one group per category, then your USER PRESETS. A click on a group title opens or closes it (the number at its right is the number of presets in it); TOP Sounds, your user presets and the group of the current preset start open. A coloured letter before each name gives its engine (A in teal, B in amber; AB in both colours = the two engines mixed). A dot after the preset name means the sound has been modified since it was loaded. The tables in this manual list the presets by category, separately for each engine.

Preset browser (beta 0.3, web app and plugin). A panel at the right of the synth lists the sound types with their number of sounds — TOP Sounds first (OktoPierre’s selection of his best sounds; beta 0.7: exactly his 22, no Source Phase sound among them), then the categories, Source Phase last (beta 0.7) — and the sounds of the type selected: one click plays a sound; ★ marks a TOP sound and a number OktoPierre’s vote in his listening test. The arrow keys choose too (beta 0.5): ↑ ↓ play the sound above or below, ← → show the category above or below (in the plugin after a click in the browser, so that it has the keyboard; with the browser hidden they step through the presets like ◀ ▶). FACTORY / USER PRESETS (USER BANK in the plugin) tabs; HIDE ▸ folds it away and a click on the preset name brings it back. In the web app it only selects sounds; the plugin also saves and arranges them (see the plugin section).

Web app. The buttons at the right of the status bar manage your own sounds:

• Save preset asks for a name (up to 40 characters) and stores the sound in this browser’s storage, in the USER PRESETS group.
• Export downloads the current sound as a .json file named after the preset — the way to back up sounds or move them to another computer.
• Import loads such a file, adds it to the user presets and selects it. Parameters missing from the file take their default value.
• To delete a user preset, open the list, hover over it and click ✕ (you are asked to confirm). Factory presets cannot be deleted.
• The app always opens on A · Warm Pad; it does not bring back the sound you had when you closed it. Save or export a sound you want to keep — your user presets stay in the list.

VST3 plugin. ◀ ▶ step through the factory presets in the same order; the preset menu (a standard Windows menu) shows TOP Sounds at the top and one submenu per category, with the number of presets in it. Save your own sounds with Live: the plugin’s settings are stored in the Live Set, and you can save the device as a Live preset (or inside an Instrument Rack) to reuse it in other projects. A new instance starts on A · Warm Pad.

16Playing

Demo (web app)

The demo control next to the logo is a split button. ▶ DEMO plays the default demo straight away; the small ▾ beside it opens the list of all demo pieces and Load a MIDI file…. While a demo plays, the main button reads ■ STOP. The 8 pieces, by L. van Beethoven and J. S. Bach, are free editions from the Mutopia Project, each cut after about two minutes at the end of a phrase; the notes are played sample-accurately by a sequencer inside the audio engine.

The default demo: a tour of the presets

▶ DEMO — and Run the demo in the welcome window — plays L. van Beethoven — Appassionata (III) (1:59) as a preset tour, starting with the phrase that follows the opening chords (7.8 s into the movement): about every 4 seconds, on a change of chord in the music, the next preset is loaded — B · Clarinet → A · Noise Wash Pad → B · Mirror Sub Bass → A · Siren Sweep … — 30 presets in all, chosen by OktoPierre (beta 0.7): the sounds of his own demo selection, from B · Clarinet on, mixed one by one with his TOP Sounds, so that 17 of the 30 presets are TOP Sounds (★ in the preset browser). The sequence is fixed in the demo, so new factory presets (there are 150) do not change it. The preset browser in the header shows the name of the preset being played (with the “•” modified mark while a control moves).

• To keep every preset playable on the piece, each preset is played polyphonic (16 voices), at its normal octave, without glide and with its attack shortened to at most 80 ms; long echoes are reduced (delay mix at most 10 %, feedback at most 30 %) so the fast piece stays clear. The presets themselves are not changed.
• The demo window shows “Preset tour · n/30 · Engine A/B” (n = the preset’s place in the tour), “Now playing:” with the preset’s name, a countdown to the next preset with its name and, below them, a line of text: during the first 6 presets one of the welcome sentences below, from the 7th preset on a one-line summary of the sound, ending with “Played polyphonic at its normal octave, with short echoes.”

A word of welcome (beta 0.7). While the first 6 presets play, the demo window explains the idea of the instrument in plain words, one sentence per preset (about 4 seconds each):

1. “Welcome to OKTOWAVE. Every sound you hear is built from wavelets: tiny waves, like bricks of sound.”
2. “The synth splits each note into eleven layers, from the deep body of the sound up to its brightest shimmer.”
3. “Engine A takes a classic waveform and shapes these layers: a filter that works with wavelets instead of frequencies.”
4. “Engine B builds the sound brick by brick: its drawbars mix the layers, like an organ made of wavelets.”
5. “Wavelets also know where things happen inside the wave, so the sound can be reshaped in ways a classic filter cannot.”
6. “Glassy pads, hollow organs, deep basses: each one is a preset you can play. Enjoy the tour!”

All demo pieces

• Scenes. The other pieces animate the controls instead of changing presets: a series of scenes follows the phrases, each varying one control. The Toccata in D minor has 11 hand-written scenes (table below); the remaining pieces get scenes placed automatically at the natural pauses of the music, from a list of scene templates chosen for each piece: filter sweep, resonance, filter slope, source wave, position window, sparsity, sub oscillator, wavelet atom shape, tilt, atom patterns, drawbars, ping-pong delay, unison and a reverb finale. The window then shows the scene and engine, “Now varying:” with the parameter’s name, its live value and range, and a one-line explanation, and the animated control glows on the panel.
• Start and stop. ▶ DEMO starts the default demo; ▾ lets you pick any piece (the list opens without starting the audio). Stop with ■ STOP in the header or in the demo window, or with Esc (when the manual overlay is closed). Drag the demo window by its title bar to move it out of the way; it also shows the progress, the time and the source and licence of the music.
• Your sound is safe. When a demo ends or is stopped, the sound you had before is restored exactly; your preset and settings are not changed. During a demo the preset browser shows what is being played: during the preset tour exactly the current preset’s name; in the other demos the name of the scene’s starting preset with the “•” modified mark (the Toccata, for example, starts on “B · Octave Organ•”). Your preset name comes back when the demo stops, and changes you make on the panel during the demo are discarded.
• Play along. The keyboard, the computer keys and a MIDI keyboard keep working during the demo; the demo’s own notes also light up the on-screen keyboard.
• One window at a time. Closing or leaving the page stops the demo and the audio immediately. If the app is open in two windows or tabs of the same browser, only one plays: starting the audio in one pauses the other, whose welcome window comes back with “The synthesizer is playing in another window or tab, so this one was paused.” in place of its sub-title; Play the synth or Run the demo there takes the audio back.

Playing your own MIDI file

Choose Load a MIDI file… in the ▾ list and pick a .mid file: it is played with automatically generated scenes, like the demo pieces. Drums (MIDI channel 10) are ignored, and only the first 6 minutes are played; a file without playable notes is refused with a message.

The Toccata in D minor scenes

Demo music: MIDI files from the Mutopia Project. Sonata "Moonlight" Op. 27 No. 2 — III. Presto agitato (L. van Beethoven) is licensed CC BY-SA 2.5 — Mutopia Project, typeset by Stewart Holmes; its converted note data in the app is shared under the same licence. All other pieces are in the public domain. The full list of sources and licences is in tools/demo/SOURCE.md. The demo is part of the web app only; the VST3 plugin has no DEMO button.

On-screen keyboard and wheels

• The keyboard spans five octaves, C2 to C7 (beta 0.7: with taller keys, and taller wheels). Click a key to play it; the vertical position sets the velocity — from 35 % at the top of a key to 100 % at its bottom edge. Drag across the keys to slide from note to note.
• The wheels sit just left of the keyboard (beta 0.6). The PITCH wheel bends by up to ±2 semitones and springs back to the centre when you let go. The MOD wheel adds vibrato (up to 50 cents, through LFO 1) and stays where you leave it. Both wheels follow incoming MIDI.
• Held notes light up in the accent colour; a thin coloured strip above the keys shows the range played by the computer keyboard.

Computer keyboard (web app)

• Physical key positions are used, so the map works on any layout. On AZERTY the white keys are Q S D F G H J K L M Ù, the black keys Z E T Y U O P, the octave keys W / X, and the digit row plays without Shift. On QWERTZ, Z and Y swap places.
• Computer-keyboard notes play at a fixed velocity of 80 %. Holding a key does not repeat the note.
• Keys are ignored while the manual overlay is open, while you type in a text field, and together with Ctrl, Alt or ⌘. If the browser window loses focus, held notes are released.
• The VST3 plugin has no computer-keyboard map of its own: use Live’s Computer MIDI Keyboard (M).

MIDI

The web app uses Web MIDI, which works best in Chrome and Edge (if a browser does not offer it, the status bar says MIDI: not supported by this browser). When audio starts it asks for MIDI access, listens to all connected inputs, and lists the devices in the status bar (e.g. MIDI: KeyStep 37). Devices plugged in later are picked up automatically. In the plugin, MIDI comes from the Live track.

Without a DAW (dawless). Keyboards, sequencers and controllers connected to the computer (USB or a MIDI interface) play the web app directly, also from the online version, which is served over https as Web MIDI requires.

• Channel. The selector in the status bar chooses the input channel: MIDI ch: all (default) or one channel 1–16, for example to play the synthesizer from one track of a hardware sequencer.
• Program Change selects a preset: program 0 is the first preset of the list, program 1 the second, and so on (your user presets follow the factory presets). The status bar shows the program and the preset.
• MIDI learn. Right-click a control on the panel (knob, drawbar or selector): it lights up and the status bar asks you to move a knob or fader on your controller. The control change you move is then linked to that control and moves it over its whole range. Esc cancels, Delete unlinks the control. A control’s tooltip shows its linked CC. A learned link takes priority over the default meaning of that CC (for example CC 1). Links and the channel are kept in this browser.

Polyphony, mono, glide and unison

• VOICES sets the polyphony (1–16). At 1 the instrument is mono and legato with last-note priority — the classic lead and bass behaviour.
• GLIDE sets the glide time and the portamento selector decides when notes glide: OFF never, POLY every voice from its own previous note (polyphonic portamento, see Voice), LEGATO only while another key is held. In mono, play legato (keep the previous key held) to slide without restarting the envelopes.
• UNISON, DETUNE and STEREO thicken each note with up to four detuned copies spread in stereo.
• Pitch reference: MIDI note 69 (A4) = 440 Hz, before OCTAVE and FINE.

17The VST3 plugin in Live

The plugin runs the same sound engine with the same panel. It is a single self-contained file inside a WaveletSynthesizer.vst3 bundle folder, which also carries these manuals; nothing else needs to be installed. The same file is both a VST3 and a VST2 plug-in (a copy named WaveletSynthesizer.dll sits next to it for VST2 hosts), so the folder works in a VST3 folder and in a VST2 folder. This section lists only what is specific to the plugin.

Installation (Windows 10/11, 64-bit)

1. Download the release zip from the plug-in’s release page — the web app’s DOWNLOAD VST button and the welcome window’s Download the VST open it — and unzip it.
2. Copy the whole WaveletSynthesizer.vst3 folder — the manuals live inside it — into one of these:
   • VST3 (recommended): C:\Program Files\Common Files\VST3. vst\install.bat does it and asks for administrator rights.
   • VST2: the folder set as VST2 Plug-in Custom Folder in Live (no administrator rights needed).
3. Start Live and open Preferences → Plug-ins. Switch on Use VST3 Plug-in System Folders (VST3) or Use VST2 Plug-in Custom Folder (VST2) and click Rescan. If both are on and see the folder, Live lists the plugin twice, once as VST3 and once as VST2; use either.
4. In the browser, open Plug-ins: Wavelet Synthesizer is listed under the vendor OktoPierre. Drag it onto a MIDI track (or double-click it with a MIDI track selected).

Building the plugin from source (with the portable llvm-mingw toolchain and vst\build.bat) is described in the project’s README.md.

Differences from the web app

• No welcome window, no DEMO or DOWNLOAD VST button and no status bar. Audio runs as soon as the plugin is loaded; there are no Save / Export / Import buttons (the plugin has its PRESET BROWSER, SAVE PRESET and user bank instead, see below). In the header, the place of the web app’s DEMO button holds the window-size button, and that of DOWNLOAD VST holds SAVE PRESET.
• Tempo sync (beta 0.6, VST3 and VST2). SYNC buttons that the web app does not have: at the right of the PING-PONG DELAY heading, of the LFO 1 title and of the LFO 2 title. Lit, the delay’s TIME or the LFO’s RATE becomes a note value at the host’s tempo (120 BPM when the host gives none) and follows its tempo changes; the knob shows the note value, e.g. “1/8D”. SYNC is off by default, so older presets, states and Live Sets sound the same. See Delay tempo sync and LFO tempo sync.
• Presets. A new instance starts on A · Warm Pad. ◀ ▶ step through the 150 factory presets (those you deleted are skipped). Clicking the preset name opens the preset browser; a right-click on it opens the classic Windows menu (TOP Sounds, one submenu per category, the USER BANK, and the save / restore / delete / export commands). You can also keep sounds with Live: as a device preset, inside an Instrument Rack, or simply in the Live Set. The plugin remembers the selected preset, and whether the browser is open, with the Live Set.
• Preset browser (beta 0.3). A panel at the right of the synth (HIDE ▸ folds it away and the window gets narrower; a click on the preset name brings it back). Two tabs: FACTORY and USER BANK. The left column lists the sound types with their number of sounds — TOP Sounds first (OktoPierre’s selection), then Bass, Lead, Keys, Organ, Pad … Init and, last, Source Phase (beta 0.7) — and the right column the sounds of the type selected: one click loads a sound, and so do the arrow keys (beta 0.5) once you have clicked in the browser: ↑ ↓ the sound above or below, ← → the category above or below. A long list scrolls with the mouse wheel or with the scroll bar at its right (beta 0.4: wide enough for a trackpad — drag it, or click the bar where you want to go). Each sound shows its engine letters (A, B or AB), ★ for a TOP sound, OktoPierre’s vote (7–10) for the voted ones and ADJ for a preset you adjusted. Above the list, ▲ ▼ move the sound playing up or down in that list. Moving a sound to another category (beta 0.5): drag it onto a category of the left column — onto TOP Sounds it is added there — or right-click it: Move to ▸ a category, Back to its own category, Add to TOP Sounds / Remove from TOP Sounds. In the USER BANK tab a sound dragged onto a section moves to that section. All of it is kept in the presets file (and shared with EXPORT / IMPORT). Below: TO BANK… (save the sound to the user bank), RESTORE (the original of an adjusted factory preset), DELETE (a factory preset leaves the lists, a bank sound is deleted), EXPORT… (a copy of the presets file) and IMPORT… (beta 0.4: another presets file replaces yours, see The presets file below). Show the presets file (in Windows Explorer or the Finder), the former FILE button, is in the right-click menu of the preset name.
• SAVE PRESET (beta 0.3; beta 0.6: its own button left of MANUAL, where the web app has DOWNLOAD VST, so the preset box is wider) overwrites the preset playing. A factory preset is adjusted in place, optionally under a new name (a small window proposes its name); a sound of the user bank is replaced. RESTORE brings a factory preset’s original back.
• User bank (beta 0.3). Your own sounds, sorted in 12 sections: Bass, Lead, Keys, Organ, Pad, Strings, Brass & Wind, Pluck, Bells & Mallets, Chords, FX, Drums & Perc. It is empty at first. TO BANK… (or Save to user bank… in the menu) proposes the current name (change it if you like) and a section; the preset name then reads USER · name. Saving a name that already exists in that section asks before replacing it.
• The presets file (beta 0.3). Everything above lives in one readable text file, OKTOWAVE presets.txt, kept next to the WaveletSynthesizer.vst3 folder (so that installing a new version keeps it), shared by every instance and Live Set, and read again when you reopen Live: every factory preset (“Adjusted: yes” for those you saved, “Deleted: yes” for those you deleted), the order of every list, and the user bank; for each sound it lists the parameters that differ from the defaults. That folder must allow writing: the VST3 system folder C:\Program Files\Common Files\VST3 needs administrator rights, so the plugin says so if it cannot save there. A deleted factory preset comes back with a right-click on the preset name, Deleted presets. beta 0.7 writes the lists as “LIST ORDER v3”: a presets file of an earlier version keeps its adjusted and deleted presets and its user bank, but its lists start again from the new defaults (the 22 TOP Sounds, Source Phase last). IMPORT… (beta 0.4; also Import presets (.txt), replacing yours… in the right-click menu) reads such a file — an EXPORT… copy, a friend’s OKTOWAVE presets.txt — and, after a question, replaces your adjusted and deleted presets, list order and user bank with its own; your previous file is first kept next to it as OKTOWAVE presets (backup date time).txt (to go back, import that copy). A file without presets is refused and nothing changes.
• Mouse. Drag vertically (Shift = fine), use the mouse wheel (Shift = fine), and double-click or Ctrl + click to reset a control to its default. A knob with only a few positions (the SUB OSC’s OCT and SHAPE) moves with a short drag, and a click goes to its next value. Hovering shows the value, as in the web app.
• Playing. The on-screen keyboard and wheels work with the mouse, but there are no computer-keyboard notes: arm the track and use a MIDI keyboard or Live’s Computer MIDI Keyboard (M). Pitch bend, the mod wheel and the sustain pedal arrive from the track’s MIDI.
• Manuals. MANUAL opens the manual of the engine shown in your default web browser, from the copy stored in the bundle: Contents\Resources\manual\manual-a.html on the A · SUBTRACTIVE tab, manual-b.html on B · ADDITIVE. Each manual links to the other.
• Version. The plugin reports vendor OktoPierre and version 0.7.0 (beta 0.7); the panel shows “By OktoPierre · beta 0.7 · GPL v3”.
• Window size. The small size button next to the logo opens a menu of window sizes: 70, 80, 90, 100, 115, 130 and 150 %. The choice is saved with the Live Set.

Automation

All panel parameters can be automated, under their full names — the “host” names given in the control tables of this manual, for example Cutoff, Envelope Amount, Source Wave, Sparsity. In Live, pick them in the track’s automation chooser, or click the plugin’s Configure button and move a control to add it to the device view.

beta 0.6: the parameters of the first LFO are now named LFO 1 Shape, LFO 1 Rate, LFO 1 to Pitch, LFO 1 to Cutoff, LFO 1 to Position and LFO 1 to Wavelet Morph (they keep their numbers, so saved automation still finds them), and the tempo sync adds Delay Sync, Delay Time (sync), LFO 1 Sync, LFO 1 Rate (sync), LFO 2 Sync and LFO 2 Rate (sync).

18Tips & troubleshooting

19Glossary

Aliasing

Harsh, inharmonic tones produced when a digital oscillator tries to play harmonics above half the sample rate; they fold back into the audible range.

Atom

One shifted, scaled copy of the wavelet, with a size (coefficient). One cycle of 2048 samples is described by 2047 atoms plus its average.

Band / scale

All atoms of the same size. Band b has 2^b−1 atoms; band 1 spans the whole cycle, band 11 has 1024 tiny atoms. One band ≈ one octave of harmonics.

Band-limited table

A copy of each band layer with harmonics above a limit removed. The engine picks the copy that fits the note, so high notes do not alias.

Daubechies (DBn)

A family of wavelets by Ingrid Daubechies (1988). DBn uses 2n filter taps; longer ones are smoother and separate the octaves more cleanly.

DWT

Discrete wavelet transform. Here it is periodic (the cycle wraps around) and orthogonal (no atom duplicates another), so the 11 band layers add up exactly to the original cycle and the energy is preserved.

Haar

The oldest and simplest wavelet (1910): a square step up then down. Its atoms are piecewise constant, which gives stepped, digital-sounding results.

Infrasound

Frequencies below about 20 Hz: under the range of hearing, but felt through large speakers.

Key tracking

How much the filter cutoff follows the played note.

Legato

Playing a new note before releasing the previous one; in mono mode the envelopes then continue instead of restarting.

Mother wavelet (ψ)

The basic shape from which all atoms are made by shifting and scaling. The ATOM SHAPE display shows it.

Pitch-synchronous

Analysed and played one waveform cycle at a time, locked to the note’s period. Bands therefore move with the pitch: band 5 is always about harmonics 9–16.

PolyBLEP

A light-weight way of removing the aliasing of the sharp edges of a digital square or sawtooth wave; used by the sub oscillator’s square.

Portamento (glide)

A smooth slide of pitch from one note to the next instead of a jump. With polyphonic portamento, every voice of a chord glides from its own previous note.

Scalogram

A picture of a wavelet transform: time across, scale (band) up, brightness = atom size. The wavelet counterpart of a spectrogram.

Sparsity

Keeping only the largest atoms and zeroing the rest — the principle behind wavelet-based lossy compression such as JPEG 2000.

Sub oscillator

An extra oscillator one or two octaves below the played note, used to add low end.

Subtractive synthesis

Starting from a harmonically rich wave and removing (filtering) parts of it — here, removing bands of atoms.

Symlet, Coiflet

Variants of Daubechies wavelets designed to be nearly symmetric, which gives more balanced, less lopsided edges.

Taps

The length of the wavelet’s digital filter. More taps = smoother atom, better octave separation, less precise placement in time.

Tempo sync

Setting a time or a speed as a note value of the song’s tempo instead of seconds or hertz: 1/8 = an eighth note, T = triplet (2/3 of the note), D = dotted (1.5 ×). In the plug-in, the SYNC buttons of the delay and of both LFOs.

Unison

Several slightly detuned copies of each note played together for a thicker sound.

Wavelet

A short, wave-like shape with zero average that starts and ends quietly. Stretched and shifted copies of it are the building blocks used to describe a signal.

20About & release history

Wavelet Synthesizer — created by OktoPierre (Pierre Arnaud GUYOT). Version beta 0.7 (2026-10-18).

• The panel shows “By OktoPierre · beta 0.7 · GPL v3” in small print at the bottom left, under the VOICE section (beta 0.7), in the web app and in the plugin.
• The VST3 plugin is published under the vendor name OktoPierre — that is where Ableton Live lists it in its browser — with version number 0.7.0.
• These manuals are generated from the instrument’s own parameter list and sound engine (tools/manual/), so ranges, defaults and presets always match this version.
• How to cite: Guyot, P. A. (2026). Wavelet Synthesizer [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.23431760 (Zenodo record).

Acknowledgments

Thanks to Jeremi Palmigani (aka Saoul Machine) and Flo (aka Go Tiger) for testing the synth and providing valuable feedback and improvement suggestions.

Thanks to Arnaud Grignard for the technical review of the concept.

Release history

The complete history is kept in RELEASE_HISTORY.md in the project folder. Current release: