flutter_audio_fx 0.4.0

Real-time voice effects for Flutter powered by a Rust DSP core: auto-tune, pitch shift, noise suppression, EQ, reverb, de-esser and more.

flutter_audio_fx

Studio-grade, real-time voice effects for Flutter — powered by a Rust DSP core.
Pitch shift, auto-tune, noise suppression, EQ, reverb and 14 effects total, with zero allocations on the audio thread.

Quick start · Effects · Presets · Platform setup · Performance · How it works

---

✨ Why flutter_audio_fx?

• 🎛️ 14 production effects. Auto-tune, pitch shift with formant preservation, RNNoise noise suppression, a 10-band EQ, dynamics, space, modulation and vocal polish, all chainable in any order.
• ⚡ Real-time and glitch-free. The audio callback never allocates, locks or does I/O. Knob changes are lock-free atomics that are smoothed per sample, so sliders don't produce zipper noise or clicks.
• 🦀 Rust DSP core with no toolchain needed. Precompiled binaries for every platform ship with the package. Add the dependency and run.
• 🎚️ Live chain editing. Reorder, toggle and retune effects while audio is running. Reverb tails keep ringing and the denoiser stays warmed up.
• 📁 Offline rendering. Process WAV files faster than real time on a worker isolate, with progress callbacks.
• 🎙️ Tap recording. Write the raw mic signal and/or the processed output to WAV while you monitor live.
• 📊 Visualizer widgets. Spectrum analyzer, waveform and a pitch/tuner indicator, ready to drop in.
• 🎭 11 built-in presets. T-Pain, Radio Host, Chipmunk, Podcast, Telephone, Robot and more.

📦 Installation

dependencies:
  flutter_audio_fx: ^0.4.0

Or track the repository directly:

dependencies:
  flutter_audio_fx:
    git:
      url: https://github.com/rickyirfandi/flutter_audio_fx.git
      ref: v0.4.0

Then follow Platform setup to enable microphone access.

🚀 Quick start

import 'package:flutter_audio_fx/flutter_audio_fx.dart';

final engine = AudioFxEngine();
await engine.init();

engine.setChain([
  NoiseSuppress(strength: 0.8),
  AutoTune(
    key: MusicalKey.C,
    scale: MusicalScale.major,
    correctionSpeed: 0.0, // hard snap: the classic T-Pain sound
  ),
  DeEsser(),
  Reverb(roomSize: 0.5, mix: 0.3),
  Limiter(ceilingDb: -1.0),
]);

await engine.startMic(); // mic → effects → speaker

// Tweak any parameter live; changes are smoothed, never clicky.
engine.updateParam(1, 'speed', 0.5);
engine.toggleEffect(3, false);

await engine.stop();

🎛️ Effects

Category	Effect	Highlights
Cleanup	NoiseGate	Threshold, attack, release, hold, hysteresis and range
	NoiseSuppress	RNNoise (pure-Rust port) neural denoiser
	DeEsser	Split-band sibilance control with an adjustable crossover
Pitch	PitchShift	Phase vocoder, ±12 semitones, formant preservation
	AutoTune	YIN detection, 12 keys, 7 scales, retune speed, humanize
Tone	Equalizer	10-band parametric EQ
	Exciter	Adds high-frequency harmonics for presence and air
	Distortion	Soft, hard, tanh and bitcrush modes, 4x oversampled
Dynamics	Compressor	Soft knee, makeup gain, sidechain high-pass
	Limiter	5 ms lookahead brick-wall limiter
Space	Reverb	Freeverb with pre-delay and damping
	DelayEffect	Up to 2 s with click-free time changes
Modulation	Chorus	Rate, depth and mix
	Doubler	Two modulated voices for double-tracked vocals

Every effect is a plain Dart object. Construct it with named parameters, then change it later with engine.updateParam(index, name, value).

🎭 Presets

engine.setChain(BuiltInPresets.tpain);

Preset	Sound
tpain	Hard auto-tune + compression
gentleAutoTune	Subtle pitch correction
radioHost	Clean broadcast voice
podcast	Professional voice cleanup
chipmunk	+1 octave pitch shift
deepVoice	Deep bass voice
lofi	Bitcrush + reverb aesthetic
karaoke	Reverb + delay for singing
telephone	Vintage phone filter
robot	Robotic vocoder effect
echoChamber	Heavy echo + reverb

BuiltInPresets.all lists every preset with an id, name, emoji and description, which is useful for building a picker UI.

🎙️ Recording

Tap the raw and/or processed signal to WAV while monitoring live. The files are written on a worker thread, never on the audio thread.

await engine.startMicWithRecording(
  rawOutputPath: '/path/raw.wav',
  processedOutputPath: '/path/processed.wav', // optional
);
// ...
await engine.stop();

📁 Offline file processing

await engine.processFile(
  inputPath: 'recording.wav',
  outputPath: 'processed.wav',
  onProgress: (p) => print('${(p * 100).toStringAsFixed(0)}%'),
);

Rendering runs on a worker isolate, so the UI stays responsive. Multi-channel input is downmixed to mono (the chain is mono) and written as mono WAV. previewFile(inputPath: ...) plays a file through the chain in real time.

MP3 export is reserved for a future release. AudioFormat.mp3() throws UnsupportedError.

📊 Visualizers

SpectrumVisualizer(stream: engine.spectrumStream, barCount: 32, height: 120)
WaveformVisualizer(stream: engine.waveformStream, height: 100)
PitchIndicator(stream: engine.pitchStream)

The raw data is also available as streams: spectrumStream, waveformStream, pitchStream and levelStream.

🔧 Platform setup

Platform	Minimum	Setup
Android	API 26	RECORD_AUDIO is merged from the plugin manifest; request it at runtime (e.g. with permission_handler).
iOS	13.0	Add NSMicrophoneUsageDescription to ios/Runner/Info.plist.
macOS	10.14	Add NSMicrophoneUsageDescription to Info.plist and the com.apple.security.device.audio-input entitlement.
Windows	10	None.
Linux	x64	ALSA (libasound2) at runtime.

⚡ Performance

Metric	Target
Audio callback time	< 3 ms per buffer
CPU (all effects on, one channel)	< 20% of a mid-range core
Memory	< 50 MB total
Chain latency (non-FFT effects)	< 10 ms

Some effects add their own delay. engine.totalLatencyMs reports the enabled effects' total plus the audio buffer latency:

Effect	Latency @ 48 kHz
Pitch Shift	1536 samples (32 ms)
Auto-Tune	1536 samples (32 ms)
Noise Suppression	480 samples (10 ms)
Limiter	240 samples (5 ms)
Distortion	16 samples (0.3 ms)
All others	0

Pitch Shift and Auto-Tune keep their delay constant even at zero shift, so correction engaging and disengaging never clicks.

🧠 How it works

Flutter (Dart)
  ├── AudioFxEngine            lifecycle, chain editing, visualizer polling
  ├── AudioEffect classes      typed, serializable effect configs
  └── Visualizer widgets
                │
                │  dart:ffi · flat C ABI (rust/src/api.rs)
                ▼
Rust DSP core
  ├── effects/                 14 effects implementing one AudioEffect trait
  ├── graph/                   EffectSlot, AtomicF32, SmoothedParam
  ├── analysis/                cached-FFT spectrum, YIN pitch detector
  ├── engine/runtime.rs        cpal streams, ArcSwap chain, worker threads
  └── io/file_io.rs            WAV reader/writer (hound)

On every buffer (≈ 2.7 ms at 128 samples / 48 kHz), the audio callback:

1. Loads the current chain with one atomic ArcSwap read.
2. Applies queued parameter changes from a lock-free SPSC queue.
3. Runs each effect in place over pre-allocated buffers.
4. Hands samples to worker threads (recording, pitch detection) via ring buffers.

No heap allocation, no blocking locks, no syscalls.

🛠️ Building from source

Package users don't need this: CI builds binaries for every platform on each version tag. For contributors working on the Rust core in rust/:

make android   # → android/src/main/jniLibs/<abi>/libflutter_audio_fx_core.so
make ios       # → ios/flutter_audio_fx_core.xcframework
make macos     # → macos/libflutter_audio_fx_core.a (universal)
make linux     # → linux/libflutter_audio_fx_core.so
make windows   # → windows/flutter_audio_fx_core.dll

cd rust && cargo test   # DSP test suite
flutter test            # Dart tests

Requires Rust ≥ 1.83, plus cargo-ndk and NDK r26+ for Android and Xcode 15+ for Apple platforms.

🤝 Contributing

Issues and pull requests are welcome at github.com/rickyirfandi/flutter_audio_fx.

Adding an effect means implementing the AudioEffect trait under rust/src/effects/, registering it in EffectType and the chain builder, then mirroring it with an AudioEffect subclass in lib/src/effects/. Please keep all audio-thread code allocation-free and lock-free.

📄 License

MIT. See LICENSE.