faust_min 0.7.2
faust_min: ^0.7.2 copied to clipboard
High-fidelity physical modeling music synthesis engine powered by Faust, featuring 50+ dynamic instruments, polyphonic support, and an advanced UML multi-notation sequencer.
0.7.2 #
- Chinese Ensemble Physical Models & Calibrations:
- Guzheng (ID 39): Tuned 4-strike plucked articulation engine with quadratic touch response, nonlinear tanh soft-saturation to eliminate square-wave clipping, and pitch stability.
- Erhu (ID 40): Tuned bowed physical string model with python-skin membrane resonances (950 Hz, 1550 Hz, 2800 Hz) and body acoustics with robust fundamental projection across D4-D6.
- Dizi (ID 53): Re-architected transverse bamboo flute waveguide onto the calibrated
fm.libcore for exact acoustic period tracking, preserving the authentic Dimo reed tissue membrane buzzing model and resonator. - Chou Gong (ID 35) & Dagu (ID 43): Integrated traditional pitch-glide Chinese opera gong and grand ceremonial drum.
- Sequence Orchestrator & Multi-Style Validation: Added Traditional Chinese Ensemble (High Mountain & Flowing Water / 高山流水) preset test suite (Selection 14 in
test_sequence).
0.7.1 #
- Flute low-register crack fixed (3rd-mode lock eliminated): The flute's two-tap waveguide with a cubic saturator had loop gain > 1 on the 3rd bore mode, so notes in the low register latched onto it instead of the fundamental (2.4-2.9x pitch jumps with a weak 3rd harmonic - the audible "crack"). The shared
fm.libmodel was reworked: single feedback tap, boundedma.tanhvalve, amode_cutofflowpass that tracks the calibrated frequency so mode selectivity is frequency-invariant, and a softerjet_kvalve knee. Verified across 147 notes at 5-cent spacing from 147-220 Hz: 0 mode locks. - Bansuri deduplicated onto the shared model:
bansuri.dspno longer carries its own copy of the waveguide; it importsfm.liband calls the samefluteModel, retaining its own embouchure split, turbulence injector, external air path, and wall-loss character. The fix therefore applies to both wooden flute instruments, and sitar/sarod/tanpura still compile against the same library. - Full-range pitch calibration re-solved: The
calibrationparameter is now a genuine pitch trim solved by closed-loop measurement against the real engine (secant + bounded scan + densification to 25-cent spacing). Flute: 181 rows, 177 within 25 cents. Bansuri: 133 rows, all 133 within 25 cents. Thecal -> pitchslope is ~1.0 at low frequency and 1.4-2.0 near 1.4 kHz, so a fixed-point solver oscillates - that is why the values were measured, not derived. - Flute & bansuri gain re-calibrated: The new waveguide is 6-17x quieter than the model the previous gains were solved against, so all gain values and both DSP multipliers were stale. Re-ran the
fixgain.mdprocedure at target 0.5. Flute energy spread across the range tightened from 2.9x to 1.18x, bansuri to 1.22x; no clipping (max peak 0.57/0.71). Pitch calibration re-verified afterwards and is unaffected, ascalibrationis gain-independent. - Flute low-range coverage extended to 146.83 Hz: The LUT and calibration grid now reach a full octave below the previous 164.81 Hz floor.
- Velocity response linearised:
sqrt(E)/vis now constant (0.66-0.76) across the full velocity range for the flute, instead of a 24x level cliff between v=0.8 and v=1.0. - Flute triggers are now deterministic: the flute (instrument ID 10) always re-articulates the gate, even when played legato. Previously, leftover bore state made identical repeated triggers randomly either lock the fundamental or burst into overtones. Note this means the flute no longer produces a true legato (seamless) repeat - each note gets a fresh attack, which is also how a real flute behaves.
- Known limitation: flute 1458.8 Hz and 1480.0 Hz have no calibration value that brings them in tune (best ~+29 cents). At ~1.45 kHz the bore delay is only ~14 samples, so the loop's mode structure is compressed there. Everything below ~1400 Hz is unaffected.
0.7.0 #
- Mixer weight mode API: New
setWeightMode()exposes the mixer's dual-mode weight system. Start in STATIC mode to set initial track weights at registration time, then switch to DYNAMIC mode for real-time weight updates (e.g., live mixing, automation). DYNAMIC mode auto-normalises track weights so their sum stays at 1.0, preventing clipping when tracks are added or removed on the fly. - Mixer FaustMixer API docs overhaul: Complete rewrite of all
FaustMixerDart-side documentation — class-level overview, per-method descriptions with parameter ranges, return values, and cross-references to related APIs. Fixed copy-paste error ongetTrackMute(was documented as mute, not query). Also documentedInstrumentMapper,FaustInstrument,FaustEngine, andSequenceOrchestrator. - Level-monitoring callback: New ergonomic
FaustMixer.onWaveformsetter receives per-block RMS/peak of the final mix (auto-manages the native callable). The low-levelregisterWaveformCallbackis deprecated. - FaustMixer weight mode C++ fix:
setWeightMode()now properly guards against race conditions with a mutex lock, andsetTrackWeight()applies the correct per-track weight when DYNAMIC mode is active. - normalise_gains.py refactor: Generalised from single-file Bansuri script to a reusable function that accepts any CSV path.
0.6.0 #
- FaustMixer track controls: The mixer core was rebuilt around a data-oriented
MixerTrackclass, exposing full per-track control from Dart —muteTrack()/unmuteTrack()/ batchunmuteTracks(List<int>)(atomic, sub-millisecond multi-track sync) andgetTrackMute()state query. Muted tracks skip DSP rendering entirely (0% CPU while silenced). - Per-track FX: Each mixer track carries its own Faust FX chain — reverb sends (
setTrackReverbSend), echo (setTrackEcho), and 3-band EQ (setTrackEQ/setTrackMid) with per-effect bypass switches (setTrackBypassEQ,setTrackBypassEcho). Tracks created beforeinit()get their FX armed automatically on start. - Per-track envelopes & AGC: Time-stamped envelope automation (
setTrackEnvelope) with linear, exponential, and S-curve interpolation, plus automatic gain control (AGC) per track for consistent levels. - New track types:
InstrumentTrack(Faust DSP instruments via the worker pool),FileTrack(WAV streaming through a lock-free SPSC ring buffer), andMemoryTrack(zero-I/O in-memory PCM playback). - Thread-safety hardening: Mixer registry switched to a recursive mutex so control calls can safely nest from callbacks.
- API ergonomics:
FaustInstrument.create()sample-rate argument is now optional (defaults to the mixer's sample rate).
0.5.7 #
- Bass gain recalibrated to 0.3: Re-ran fix_instrument_gain.py with --target 0.3 (drone/comping level instead of melody 0.5). Max gain 0.52 folded into DSP. Verified: 82Hz=0.329, 196Hz=0.298, 392Hz=0.329.
0.5.6 #
- Debug-log spam fix:
DEBUG_INSTRUMENTis no longer compiled in by CMake — it's now a#defineinFaustInstrument.cpp(default 0). The per-note[LUT]/[NOTEON]printf spam (100k+ lines per song) is gone from all release builds. Also fixed the corresponding pre-release check inpublish.md. - Snare pitch + gain calibration: Removed the stale
freq_nudge(±15% detune) fromsnare_calibration.csv— the snare's shell tone is a pureos.osc, so it needed no pitch correction. The nudge was also shifting LUT gain lookups off the CSV grid, which madef=100impossible to calibrate. Snare gain re-calibrated to the 0.5 target withmax_gainfolded into the DSP. - Bass / Cowbell gain calibration: Completed steps 5–7 of the gain-fix procedure (normalize CSV, fold
max_gaininto DSP, revert slider to 0–1). Bass calibrated to 0.5, cowbell to 0.3.
0.5.5 #
- Song-switch crash fix (Android/iOS): Sequences now use
std::shared_ptrlifetime shared between Dart and the audio thread, with an FFI-sideUMLSequence*registry holding the Dart-owned reference. Switching songs no longerdeletes a sequence while the audio thread is still rendering it (no morepthread_mutex_lock: abort/ destroyed-mutex crashes). - Singleton orchestrator API:
FaustEngine.getOrchestrator()andFaustEngine.getMixer()now expose the sequence orchestrator / mixer as true singletons. - Deprecated legacy endpoints:
SequenceOrchestrator()constructor anddispose(), and nativeorchestrator_create/orchestrator_destroy, are deprecated (still functional) — will be removed in a future build. PreferFaustEngine.getOrchestrator()/orchestrator_get_instance()andclearSequences(). - New native APIs:
orchestrator_get_instance(),orchestrator_clear_sequence(),orchestrator_clear_sequences(). - Memory leak fix:
loadSong()no longer leaks the sequence allocation.
0.5.4 #
- UML parser regex crash fix: The tokenizer regex used
\~,\_,\>escapes that LLVM libc++ (std::regex) rejects — this crashed any UML sequence parse (song load, inlineaddSequence, etc.) on Android and iOS. Escapes removed; parsing behavior unchanged.
0.5.3 #
- FaustEngine.init() Skip-Logic: Now skips re-extracting DSP assets when the target file's size and mtime already match the bundled source asset (byte-identical fallback), making init near-instant on repeat launches.
- Init in Examples & Tests:
FaustEngine.init()added to example apps and Flutter tests, and documented in READMEs. - Native Track Muting:
muteTrack/unmuteTrack/isTrackMutedmixer APIs. - ChurchBell Fix: Reduced hum subharmonic gain (modeGains(0) 1.0→0.4) so the prime/fundamental dominates pitch; fixed header capitalization so the generated DSP is
FaustChurchbellDSP.hpp(case-sensitive iOS build).
0.5.2 #
- Frequency Calibration for All Instruments: Per-instrument pitch calibration tables (bansuri, sitar, sarod, tanpura) with cents-model correction and
pitchcorrect.pyautomation for pitch drift. - Volume Calibration for All Instruments: Gain calibration to target levels across instruments, including bansuri gain calibration to 0.5 target and piano/sitar DSP gain tuning.
- Instrument Fixes: Fixed pitch drift (violin, flute), improved bansuri breath attack, tanpura DSP updates, snare hiss parameter, and general DSP physical-model refinements across all instruments.
0.5.1 #
- Packaging fix: Include vendored
deps/oboein published package for Android builds.
0.5.0 #
- 50+ Calibrated Instruments: All instruments tuned and calibrated across multiple frequencies with measured energy (RMS, peak, Goertzel) per-instrument measurement windows.
- Kick Drum Gain Recalibrated: 37.5→15 multiplier, E=0.32 P=0.84 target.
- Tibetan Percussion Balance: Chougong, LagNga, Ngachen gain tuning for balanced energy levels.
- InstrumentMapper Measurement Tables: Per-instrument
getMeasureStart()/getMeasureDuration()for accurate energy measurement windows. - Goertzel Energy in test_instruments:
--rendermode now reports Goertzel energy at fundamental frequency. - UML Sequencer & Mixer Updates: Envelope support for track volume control.
- cmake OBJECT_DEPENDS Fix:
FaustInstrument.cppdepends on all generated hpp headers for correct rebuild propagation. - computeRMS Stereo Fix: Corrected stereo energy division (numFrames*2).
0.4.0 #
- Ghatam Instrument (ID 50): Clay pot percussion with rich midrange resonance.
- Mridangam Instrument (ID 49): South Indian double-headed drum physical model.
- Dayan DSP Redesign: Dual-bank composite membrane model — Syahi (perfect harmonic 1×–5×) + Maidan (Bessel inharmonic 1.000, 1.593, 2.135, 2.295, 2.653), with per-stroke mix ratios.
- Violin Physical Model: Safety clamps removed for clean gate-off, gain multiplier 1.1, strike=2 bowVelocity 0.15 for ≥392 Hz.
- FAUST_DISABLE_INTERPRETER: New build macro; iOS (only) cannot use interpreter mode. Android guarded similarly. All other platforms support both StaticCompiled and InterpretedByte.
- Vendored FAUST Headers: Architecture headers copied to
src/faust_arch/for portable builds without FAUST SDK. - iOS/macOS Forwarders: Renamed from
.cto.mmfor Objective-C++ compilation; includes path fixed. - CMake Portability:
libfaustlinked viafind_libraryinstead of hardcoded path; system libfaust found automatically; test binaries use vendored headers. - Podspec Metadata: Proper homepage, author, description; iOS definines
FAUST_DISABLE_INTERPRETER. - Case-Sensitive HPP Fix: Removed duplicate
FaustChurchbellDSP.hpp/FaustMaster_reverbDSP.hppconflicting withFaustChurchBellDSP.hpp/FaustMasterReverbDSP.hpp. - analysis_options.yaml: Added to root for pub.dev compliance.
0.3.1 #
- Tibetan Bowl Instrument: Added
tibetanbowlphysical model (ID 47), assigned to idiophone → melody track. - Dual LagNga Rhythm: Interleaved 111 Hz and 222 Hz LagNga oscillators at BPM=60 for ~1 hit/second.
- UML Cell Count Fix: Standalone
~/^operators now create ContinuityDot tokens at current gridIndex, fixing inter-instrument desync. - Instrument Classification: Added
isMembraneophone()/isIdiophone()for track routing instead of flat percussion flag. - Mixer Weight Fix:
setInstrumentWeight()now properly updatesTrackInstrument::instrumentWeightin mixer track registry. - Amplitude Glide Rename:
velocityGlide→amplitudeGlide(correct naming for amplitude parameter). - Base Frequency API: Exposed
sequence_get_basefreq/sequence_set_basefreqin Dart FFI. - Dead Code Removal: Removed unused
mAssignedWeightfrom FaustInstrument. - Dartdoc Coverage: Added documentation comments to all public constructors, methods, and getters.
0.3.0 #
- Bansuri Breath Attack: Added post-waveguide pink noise burst with rising-edge detection and 80ms decaying envelope for realistic breath attacks at note start.
- Bansuri Glide Fix: Resolved pitch glitches during meend transitions using K=3 Inverse Distance Weighting (IDW) LUT blending and DSP
glideparameter synchronization. - Chak Click Fix: Implemented early noteOff (30ms lookahead) in UMLParser to prevent residual waveguide energy from causing pop artifacts on fast note transitions.
- Memory Leak Fixes: All test files now properly deallocate heap-allocated
UMLSequenceandFaustInstrumentobjects. Verified clean across all test suites with ASan. - DSP Stability: Added
si.smoosmoothing to bansuri gain parameter to prevent zipper noise during parameter updates. - Native Sequencer: Sample-accurate native C++ engine for layered audio patterns — refined timing and reduced drift.
- Improved Strike Realism: Enhanced noise-burst contact simulation for percussive instruments (Bell, Bowl, Dayan, Bayan).
0.2.0 #
- Added FaustBell: High-fidelity physical modeling synthesis of a tuned meditative bell.
- Added FaustBowl: High-fidelity Tibetan Singing Bowl with strike and sustain (rubbing) physics.
- Added Native Audio Sequencer: Sample-accurate native C++ engine for layered audio patterns.
- Implemented Duration-based decay: Precisely control ring length in seconds for Bell and Bowl.
- Improved Strike Realism: Added noise-burst contact simulation for percussive instruments.
- Expanded native bindings to include all percussion and melodic instruments (Dayan, Bayan, Kick, Snare, HiHat, Tom, Ride, Sitar).
0.1.0 #
- Initial release.
- High-fidelity Faust Flute physical model.
- Raw PCM data rendering interface via FFI.
- Support for Android, iOS, Linux, MacOS, and Windows.
- MIT License.