README from GitHub, GPL-3.0
Felucca MPE
Started on 13 October 2026.
MPE (MIDI Polyphonic Expression) for the M-Vave FM-1, plus Voice LFO (VLFO) and DRIFT: chords that move and breathe, on any keyboard. By Stanley Gurvich, Emerald Moon Labs. Built on Felucca 1.6.1 by Leo Kuroshita, Hügelton Instruments (GPL-3.0-only).
Branch: main is Felucca MPE 1.6.1.
[!WARNING] Tested only in the emulator, not on a real FM-1. Everything so far ran off the device: played in the browser emulator (the same C source compiled to WebAssembly, so the synth engines and the MIDI handling are the real code) and checked by Felucca's host tests. What has not been tried is the FM-1 itself: flashing, USB MIDI on the device, and running on its CPU. The 1.6.1 build has also not been re-checked against the device's size and audio-timing budgets (the 1.5.1 numbers are below). If you flash it, you do so at your own risk: keep the official firmware at hand (M-VAVE's updater, or Felucca's web installer's Return to official V15).
Firmware file
prebuilt/felucca-mpe-1.6.1.fwsc: the package for the FM-1, built from this repository's
main (commit e8c7f1a) with ./build.sh, JieLi's toolchain and the AC79 SDK files the build checks.
[!CAUTION] This file has not been tested on real hardware. It has never been installed or run on an FM-1. It is a development build with Felucca's package identity (
FM-1_900), so installers and editors see it as Felucca. Install it only if you can recover the FM-1 (M-VAVE's updater, or Felucca's web installer's Return to official V15).
SHA-256: cb9ca992036fa6f190f78b09f83e5ab138e6e82091d2d347f2789a306fa18bfe
Played with real MPE controllers
Played in the emulator with a ROLI Seaboard RISE 49 and an Embodme Erae Touch (Mk 1): it works, and it feels like any other MPE synth.
Voice LFO (VLFO) and DRIFT: every voice its own movement, no MPE controller needed
Two new modulation sources give each note of a chord a life of its own. They work on any MIDI IN setting, on the FM-1's own keys and on its sequencer, and cost nothing while no slot uses them.
- Voice LFO (VLFO): the track's LFO, read at a different point of its cycle for each voice (the eight voice slots sit 45° apart). With the plain LFO a chord's notes all move together; with VLFO they move out of step. VLFO → PITCH, AMT 1–2 and a slow RATE: an ensemble, a gentle chorus made of the notes themselves. VLFO → CUT: a chord that shimmers note by note. It follows the LFO's RATE, SYNC, WAVE, FADE and the matrix's DEPTH. It is always bipolar (POL UNI doesn't apply), and on WAVE S&H every voice gets the same value.
- DRIFT: a slow random wander, different for every voice, with the LFO left free for something else. Each voice heads for a new random target about every 3 seconds and glides there, each at its own moment, so the notes drift apart and back without ever repeating. DRIFT → PITCH, AMT 1 (about ±15 cents) or 2 (about ±30): the drift of an old analog synth. DRIFT → CUT: a filter that breathes on every note.
Set them on the MOD page: a slot's SRC to VLFO or DRIFT, its DST to PITCH, CUT, SHP or AMP (per note), and AMT to taste.
What it does
Felucca turns the FM-1 into a deep little synth and groovebox. Felucca MPE lets you play it from an MPE controller, with every finger its own voice and its own expression:
- Each note bends on its own. Slide one finger and only that note glides; the rest of the chord stays where it is. ±48 semitones by default (the MPE standard), up to ±96. Two fingers on the same key are two separate voices.
- Press and Slide shape each note. Two modulation sources, PRES (how hard you press) and TIMB (sliding along the key), go to any per-note destination: press harder for louder (AMP), slide to open the filter (CUT), press for vibrato (PITCH), or reshape the tone (SHP). Each finger moves only its own note.
- Chords that breathe, on any keyboard: Voice LFO (VLFO) and DRIFT (above).
- A bigger bend range everywhere: up to ±96 semitones on every MIDI IN setting (Felucca: ±24).
- Everything else is Felucca 1.6.1, unchanged. The same sounds, sequencer, effects and projects.
Quick start
- Your controller: MPE on, lower zone, 15 member channels (channel 1 the master, 2 to 16 the notes), pitch-bend range 48. That is the usual MPE default; on the Seaboard RISE 49 and the Erae Touch it is set in the controller's own editor software.
- The FM-1: MENU > MIDI > MIDI IN: MPE (the last value, after TRACK). MPE plays the selected track.
- The track: EDIT > VOICE: POLY, and an engine with plenty of voices (see below).
- Expression: on the MOD page, set a slot's SRC to PRES or TIMB and its DST to AMP, CUT, SHP or PITCH; AMT sets how much. A good start: PRES → AMP (the note swells as you press) and TIMB → CUT (slide up to open the filter).
MIDI IN is saved with the project, so set it to MPE once per project. Full guide: MPE.md.
Voices: how many fingers at once
The FM-1 has 8 voices, shared by its four tracks. MPE plays one track (the selected one), so that track gets all 8 while the other tracks are silent. How many it can sound at once also depends on its engine, since each engine has a voice cap of its own:
| Engine | Voices per track | Fingers in MPE |
|---|---|---|
| ANALOG, PHASE, LOFI, SAMPLE, SLICE, TRIO, WHEEL, NOISE | 8 | up to 8 |
| FM6 | 6 | up to 6 |
| VOICE | 4 | up to 4 |
| GRAIN, PHYS | 3 | up to 3 |
| DRUM | one per lane | notes play hits, finger bends ignored |
FM6 is 6 voices: FM6 is a six-operator FM engine, and each of its voices keeps a full six-operator state in the track's engine memory; six of them fill it, so FM6 stops at 6. Play a seventh finger and one voice is taken over.
Past the limit (a 9th finger on ANALOG, a 7th on FM6, or fewer voices left because another track is sounding), the new note takes a voice over: first the oldest released note (one ringing out), then the oldest held note, never the lowest held note (your bass stays). The FM-1's own sequencer, playing other tracks at the same time, uses the same 8 voices.
Other voice modes: MONO and LEGATO play one voice that follows your latest finger (its channel and its bend); UNISON stacks voices on each finger and they bend together.
The four new modulation sources
On the MOD page (SRC, after SLEW). On a per-note destination (PITCH, CUT, SHP, AMP) every voice gets its own value; on the others (PAN, the sends, the engine knobs, RATE) they take the latest note's.
| Source | What it is | Try |
|---|---|---|
| PRES | the finger's pressure (MPE channel pressure); outside MPE, the track's aftertouch | PRES → AMP: swells. PRES → PITCH, small AMT: vibrato by pressing |
| TIMB | the finger's slide along the key (MPE CC74); 0 outside MPE | TIMB → CUT: a filter per finger |
| VLFO (Voice LFO) | the track's LFO at a different phase for each voice | VLFO → PITCH, AMT 1–2, slow RATE: ensemble. VLFO → CUT: a shimmering chord |
| DRIFT | a slow random wander per voice (a new target every ~3 s) | DRIFT → PITCH, AMT 1 (±15 cents) or 2 (±30): analog drift |
VLFO and DRIFT work on any MIDI IN setting and on the sequencer's notes, not only in MPE.
Good to know
- Projects: MIDI IN MPE is stored as value 6. Upstream Felucca does not know it (1.6.1 opens it as TRACK), so set MIDI IN back before moving a project to upstream firmware. Projects saved by the earlier 1.5.1-MPE builds stored MPE as 5, which is now Felucca's TRACK: they open as TRACK; set MPE again.
- No MPE controller? Set MIDI IN to CH1-4 and send each note on its own channel 1 to 4: each plays its own track with its own bend. With the same sound on tracks 1 to 4, that is 4-finger MPE.
- Still to come: the controller setting the zone itself (MPE's RPN 6), the upper zone, release velocity (Lift).
Building and trying it
The same as Felucca: BUILDING.md for the firmware (./build.sh) and the browser emulator
(web/emu, with Emscripten). Upstream's engines are a git submodule: clone with --recursive or run
git submodule update --init. How the MPE code is laid out and tested: MPE.md.
Credits and licence
Felucca MPE by Stanley Gurvich, Emerald Moon Labs. Felucca by Leo Kuroshita, Hügelton Instruments, with its contributors (see CREDITS.md). GPL-3.0-only, as Felucca: see LICENSE and LICENSING.md. Not affiliated with M-VAVE, ROLI or Embodme.
Changelog (one entry per increment)
| # | Version | What was added |
|---|---|---|
| 1 | v1.5.1-MPE | MIDI IN: MPE. A lower zone on the selected track: channel 1 the master, channels 2..16 a finger each with its own pitch bend (Glide), ±48 semitones by default. Two fingers on one key are two voices; a released note keeps its bend through its release. Host tests, and pitches measured in the WebAssembly build's audio. |
| 2 | v1.5.1-MPE | Bend range up to 96 semitones on every MIDI IN (upstream: 24), so a Seaboard's 48 also works in CH1-4 ("mock MPE": a finger per track). |
| 3 | v1.5.1-MPE | Fix: no hanging note when MIDI IN changes while a note is held (it hung until MIDI IN went back to CH1-4). Entering MPE drops the old routing's track bends. |
| 4 | v1.5.1-MPE | Press and Slide per finger: the matrix's PRES and TIMB sources (MOD SRC, after SLEW). Each finger's channel pressure and CC74, smoothed, on its own voice: PRES → AMP makes a finger sound as hard as it presses, TIMB → CUT a filter per finger. Outside MPE, PRES is the track's aftertouch. |
| 5 | v1.5.1-MPE | Device build checked (JieLi toolchain, upstream's tools/build.py checks and tests/target_budget.py): the per-voice MPE work left the audio interrupt's loops (shifts and a table instead of 4 divides; one helper called only while MIDI IN is MPE or a slot takes PRES / TIMB). Flash +224 B over upstream 1.5.1, RAM +512 B, audio-ISR static cost +5 % (within upstream's budget). |
| 6 | v1.5.1-MPE | VLFO: a matrix source, the track LFO at each voice slot's own phase (slot k at k × 45°). A chord's notes move out of step: ensemble, or drift at a slow RATE with a small AMT on PITCH (note) / CUT (note). Measured on a chord, PITCH AMT 6: LFO keeps the notes within 6 cents of each other, VLFO spreads them up to 158 cents. Always bipolar (POL UNI does not apply); WAVE S&H gives every voice the same value. Device: flash +76 B, RAM 0, audio ISR unchanged (it runs only while a slot takes it). |
| 7 | v1.5.1-MPE | DRIFT: a slow random wander per voice, a matrix source, the LFO left free. One global block count and one seed, hashed per voice (track × 8 + slot): a new random target every ~3 s, a straight glide, each voice at its own point of the cycle. Measured on a chord, PITCH (note) AMT 2, 12 s: each note ranges 34–58 cents on its own path. Device: flash +180 B, RAM +16 B, audio ISR ~+1 % (budget ok). Also ideas/: what else is nearly free, marked done / verified / unverified. |
| 8 | v1.6.1-MPE | Felucca 1.6.1 underneath (upstream merged in: 1.5.5 to 1.6.1, the PHASE / sample-pack licensing). MIDI IN: MPE is now the 7th value (6): upstream 1.5.5 took 5 for TRACK, so projects saved with MPE by the 1.5.1-MPE builds open as TRACK; set MPE again. The MPE work follows upstream's second core (1.6.1, MENU > SYSTEM > 2ND CORE, beta): the voices' PRES / TIMB / VLFO / DRIFT are each part's own, so two cores rendering parts at once do not share them (upstream's dual-core test now carries these sources, clean under ThreadSanitizer). Fixed in the merge: MIDI IN TRACK listened on every channel. Host tests pass; the device build was not re-measured. |
Device cost (measured, not estimated)
| Upstream 1.5.1 | MPE fork | Room | |
|---|---|---|---|
| App image (flash) | 495,588 B | 496,068 B (+480, with VLFO and DRIFT) | XIP region 581,564 B: ~84 KB free |
| RAM .data + .bss | 54,964 B | 55,492 B (+528) | of 73,728 B: ~18 KB free |
Audio ISR (fm1_alnk0_irq), target_budget cost | 38,871 | 41,218 (budget 37,502, limit +10 %) | 15 divides in loops, as upstream |
Measured on 1.5.1-MPE by building the app with JieLi's toolchain (tools/get_toolchain.sh), without the SDK (packaging
only). Not yet re-measured on 1.6.1-MPE.
Below: upstream Felucca's own README, unchanged.
Felucca
Maintenance Notice
What began as a personal experiment grew far beyond my expectations. I’m grateful to everyone who contributed, supported the project, and made music with Felucca.
I’ve decided to step back from active development to focus on creating new instruments and music.
Felucca will remain available with occasional maintenance, but new features, issue responses, and pull request reviews are not guaranteed.
Thank you for everything. I hope you’ll look forward to my next creations.
Overview
Multi-engine synthesizer firmware for the M-VAVE FM-1, version 1.6.1. Connect the FM-1 to a computer by USB, open the web installer in Chrome or Edge and press Install; no extra hardware is needed. Installing is at your own risk: M-VAVE's updater or the installer's Return to official V15 takes you back. Try it in your browser, no FM-1 needed.
- Install: web installer
- Edit: web editor (Felucca-WebApp)
- Read: the manual and the release notes are in the wiki
- Build: BUILDING.md
- Related: Felucca-Engines (the ANALOG, LOFI, TRIO, WHEEL and NOISE engines, MIT; still maintained) · felucca-bsp (the FM-1 hardware layer)
AI disclaimer
Felucca is developed with the assistance of AI coding agents (code, tests, documentation, translation, maintenance). The design, features, sound design and direction are the maintainer's or the community's, and no generative AI is used for music, icons or artwork. See On AI-Assisted Development and Responsibility.
Credits and licence
Free software: GPL-3.0-only; the Hügelton Sample Pack is CC0 1.0; the engines in firmware/engines are MIT
(Felucca-Engines); the bundled fonts and ported DSP keep their own licences
(LICENSES/, LICENSING.md). Thanks to everyone whose work and contributions are in it: CREDITS.md.
M-VAVE and FM-1 are trademarks of their respective owners. Felucca is not affiliated with or endorsed by them.
Copyright (C) 2026 Leo Kuroshita (@kurogedelic), Hügelton Instruments