⚡ Hardware SIMD-accelerated real-time audio pitch detection, SOLA/WSOLA pitch shifting, multi-track timeline mixing with stereo panning, real-time noise suppression, acoustic feature extraction, and zero-allocation DSP filters.
FastAudioProcess provides native C++ AVX2 vector processing and DSP algorithms for Java audio applications, enabling high-throughput noise suppression, pitch tracking, multi-channel timeline mixing, acoustic classification, and format conversions without Garbage Collection stalls.
import fastaudioprocess.FastAudioMixer;
import java.nio.file.Files;
import java.nio.file.Path;
public class MixerDemo {
public static void main(String[] args) throws Exception {
// 1. Initialize 44.1 kHz Stereo Mixer
FastAudioMixer mixer = new FastAudioMixer(44100, 2);
// 2. Add background audio / music bed at t=0ms (pan center, 70% volume)
float[] music = new float[44100 * 5]; // 5 seconds
mixer.addTrack(music, 0, 0.70f, 0.0f);
// 3. Layer speech audio at t=1500ms (panned slightly left, 100% volume)
float[] narrator = new float[44100 * 3]; // 3 seconds
mixer.addTrack(narrator, 1500, 1.0f, -0.25f);
// 4. Export mixdown directly as complete WAV bytes with soft-limiter protection
byte[] wavBytes = mixer.toWavBytes();
Files.write(Path.of("mixed_output.wav"), wavBytes);
System.out.printf("Exported mixed audio: %d ms%n", mixer.getDurationMs());
}
}import fastaudioprocess.FastAudioProcess;
import fastaudioprocess.FastAudioEqualizer;
import fastaudioprocess.FastAudioChunker;
import fastaudioprocess.FastFFT;
public class DSPDemo {
public static void main(String[] args) {
// 1. Generate 1 second 440Hz sine wave buffer with background noise
float[] audio = new float[44100];
for (int i = 0; i < audio.length; i++) {
audio[i] = (float) Math.sin(2.0 * Math.PI * 440.0 * i / 44100.0);
}
// 2. Real-time Spectral Subtraction Noise Suppression (FastFFT)
FastAudioProcess.suppressNoise(audio, 44100, 1.0f, 0.02f);
// 3. Acoustic Feature Analysis (Crest Factor & Periodicity)
float crest = FastAudioProcess.computeCrestFactor(audio);
float zcr = FastAudioProcess.computeZeroCrossingRate(audio);
float periodicity = FastAudioProcess.computeAutocorrelationPeriodicity(audio, 35, 160);
// 4. AVX2 SIMD Pitch Detection with DC Removal
float pitch = FastAudioProcess.detectPitchNative(audio, 44100);
System.out.printf("Pitch: %.2f Hz | Crest: %.2f | ZCR: %.3f | Periodicity: %.2f%n", pitch, crest, zcr, periodicity);
// 5. Stateful Streaming 3-Band Equalizer (Bass +3dB, Treble +2dB)
FastAudioEqualizer eq = new FastAudioEqualizer();
eq.setGains(3.0f, 0.0f, 2.0f);
eq.process(audio);
}
}- Why FastAudioProcess?
- Key Features
- Real-World Use Cases
- Architecture Overview
- Performance Benchmarks
- API Quick Reference
- Technical Demos & Benchmarks
- Installation
- Documentation
- Platform Support
- Related Projects
- License
Standard Java audio loops suffer from float boxing overhead, slow software resamplers, and JVM GC pauses that cause real-time audio crackles. FastAudioProcess solves this by:
- AVX2 SIMD Vector Acceleration — Uses 256-bit SIMD registers to process multiple float audio channels in parallel.
-
Multi-Track Timeline Mixer (
FastAudioMixer) — Sample-accurate track layering with constant-power stereo panning and analog-style soft-knee saturation limiter. -
Spectral Power Noise Suppression — Real-time in-place stationary noise cancellation (
$< 0.1\text{ ms}$ latency). - Acoustic Transient & Periodicity Extraction — Instantaneous Zero-Crossing Rate (ZCR), Crest Factor, and Autocorrelation analysis.
- Native Autocorrelation Pitch Tracking — Detects fundamental frequencies (F0) at sub-millisecond speeds.
- True WSOLA Pitch-Shifting — Echo-free musical semitone pitch modulation preserving playback duration.
- Off-Heap Direct Buffers — Operates directly on native memory buffers to prevent JVM Garbage Collection stutters.
| Feature | javax.sound.sampled | TarsosDSP | FastAudioProcess |
|---|---|---|---|
| DSP Acceleration | None (pure scalar byte loops) | Scalar Java float loops | AVX2 SIMD 256-bit Vectorized |
| Multi-Track Timeline Mixer | None | Limited / basic mixing | FastAudioMixer with Constant-Power Pan & Limiter |
| Spectral Noise Reduction | Not supported | Slow FFT / object allocations | In-place Native (< 0.1 ms latency) |
| Pitch Tracking (F0) | Not supported | Yin / MPM (~5-15 ms latency) | SIMD Autocorrelation (< 0.5 ms) |
| Heap / GC Overhead | High (buffer allocations per frame) | Moderate (heap float arrays) | Zero GC (Direct native buffer ops) |
- 🎛️ Multi-Track Timeline Mixer (
FastAudioMixer) — Sample-accurate audio layering, constant-power stereo panning (-1.0 to +1.0), and soft-knee limiter for dialogue, podcasts, and sound engines. - ⚡ Native AVX2 SIMD Acceleration — Accelerated float audio vector math for 44.1kHz / 48kHz audio streams.
- 🔇 Real-Time Noise Suppression — High-speed spectral subtraction & Wiener gain smoothing for studio voice isolation.
- 🔊 VLC-Style Volume Boost & Soft-Limiter — Volume scaling up to 300% on 16-bit PCM with continuous
tanhsoft-knee saturation limiting preventing harsh digital clipping. - ⏱️ Pitch-Invariant Time-Stretching (WSOLA) — High-fidelity Waveform Similarity Overlap-Add algorithm for speed adjustment with Hann windowing and zero Mickey-Mouse pitch distortion.
- 🎵 Clean Echo-Free Pitch-Shifting (
pitchShift) — High-quality musical semitone shifting (-12.0 to +12.0) combining WSOLA duration pre-expansion with phase-aligned resampling, completely eliminating flanging and circular echo artifacts. - 📊 Acoustic Feature Extraction — Zero-allocation Crest Factor, Zero-Crossing Rate (ZCR), and Harmonic Periodicity primitives.
- 🎤 Real-Time Pitch Tracking — High-speed autocorrelation pitch estimator for voice and instrument tuning.
- 🎛️ Zero-GC DSP Filters — Multi-channel gain, 3-band equalizer, dynamic noise gate, and FIR filters.
- 🔄 Interoperable Java Audio Bridge — Seamless conversion to and from standard
javax.sound.sampledformats.
- 🎙️ Multi-Voice Audio Dramas & Podcasts: Assemble dialogue, background music beds, and sound effects across a sample-accurate timeline with
FastAudioMixer. - 🗣️ Live Voice AI & STT Preprocessing: Eliminates microphone hiss, room fans, and background noise prior to Whisper/FastSTT.
- 🎧 VAD & Voice Feature Classifiers: Supplies fundamental acoustic metrics (Crest Factor, ZCR, Periodicity) to FastVAD.
- 🎸 Digital Audio Workstations (DAW): High-speed audio effect plugins running on FastAudioCapture streams.
- 🤖 Voice Bots & Agent Pipelines: Studio-quality voice clean-up with zero cloud API minute-fees.
- 🎮 Game Sound Engines: Multi-channel audio mixer with real-time pitch shifting for sound effects.
┌─────────────────────────────────────────────────────────────┐
│ Input Audio Tracks / Streams │
│ (Raw Floats, 16-bit PCM, WAV Files, Microphone) │
└──────────────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ FastAudioMixer │
│ • Sample-Accurate Timeline Layering (addTrack) │
│ • Constant-Power Stereo Panning law (pan) │
│ • Tanh Soft-Knee Limiter (applySoftLimiter) │
└──────────────────────────────┬──────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────────┐
│ FastAudioProcess │
│ • Spectral Noise Suppression • WSOLA Time-Stretch │
│ • AVX2 Pitch Detection • Clean Pitch-Shifting │
│ • Crest Factor, ZCR, Periodicity │
└──────────────┬───────────────────────────────┬──────────────┘
│ │
Hardware SIMD Direct Output
│ │
▼ ▼
┌──────────────────────────────┐ ┌─────────────────────────────┐
│ FastSIMD (AVX2 Engine) │ │ Interleaved Floats / WAV │
└──────────────────────────────┘ └─────────────────────────────┘
In the official JMH Benchmark, FastAudioProcess measured throughput across real-time DSP audio operations on 512-sample (16 kHz / 32 ms) frames:
Benchmark Mode Cnt Score Units
JMH_Audio.benchmarkAcousticCrestFactor thrpt 2 1,619,341 ops/s
JMH_Audio.benchmarkAcousticZCR thrpt 2 1,428,964 ops/s
JMH_Audio.benchmarkStreamingEqualizer thrpt 2 416,658 ops/s
JMH_Audio.benchmarkFastFFT thrpt 2 182,448 ops/s
JMH_Audio.benchmarkPitchDetection thrpt 2 48,836 ops/s
JMH_Audio.benchmarkSpectralDenoise thrpt 2 8,189 ops/s
Sub-Microsecond Latency: Acoustic features execute at 1.6M+ operations/sec, native AVX2 pitch tracking delivers 48,800+ detections/sec, and native FastFFT transforms run at 180,000+ transforms/sec with zero runtime GC allocations.
| Class / Method | Description | Path |
|---|---|---|
FastAudioMixer |
Multi-track timeline mixer with constant-power panning and soft limiter. | Reference |
suppressNoise(samples, rate, factor, floor) |
Real-time spectral power subtraction noise filter. | Reference |
pitchShift(input, semitones, rate) |
High-fidelity, echo-free musical pitch shifting. | Reference |
timeStretchWSOLA(input, speedRatio, rate) |
Pitch-invariant time-scale modification (WSOLA). | Reference |
applyVolumeGain(pcmData, volumeMultiplier) |
Soft-knee limiter volume boost up to 300%. | Reference |
detectPitchNative(samples, rate) |
AVX2 SIMD pitch detection using autocorrelation. | Reference |
computeCrestFactor(samples) |
Spectral Crest Factor (Peak / RMS) transient spike detector. | Reference |
computeZeroCrossingRate(samples) |
Normalized Zero-Crossing Rate (ZCR) sign-change ratio. | Reference |
computeAutocorrelationPeriodicity(samples, min, max) |
Normalized pitch harmonic periodicity ratio. | Reference |
FastAudioEqualizer |
Stateful 3-band streaming IIR equalizer. | Reference |
FastAudioChunker |
Lock-free SPSC power-of-two frame chunker. | Reference |
FastFFT |
Forward & inverse Radix-2 Fast Fourier Transform. | Reference |
| Case | Java Example | Launcher | Description |
|---|---|---|---|
| Interactive Showcase Demo | Demo.java | run-demo.bat |
End-to-end interactive demonstration of real-time audio filters, noise suppression, and pitch modulation. |
| JMH Microbenchmark Suite | Benchmark.java | run-benchmark.bat |
Formal OpenJDK JMH throughput measurements across native DSP audio kernels. |
Add the JitPack repository and the complete dependency stack to your pom.xml:
<repositories>
<repository>
<id>jitpack.io</id>
<url>https://jitpack.io</url>
</repository>
</repositories>
<dependencies>
<!-- FastAudioProcess Engine -->
<dependency>
<groupId>com.github.andrestubbe</groupId>
<artifactId>FastAudioProcess</artifactId>
<version>0.1.7</version>
</dependency>
<!-- FastSIMD Hardware Vector Acceleration Engine -->
<dependency>
<groupId>com.github.andrestubbe</groupId>
<artifactId>FastSIMD</artifactId>
<version>0.1.3</version>
</dependency>
<!-- FastCore Native Loader -->
<dependency>
<groupId>com.github.andrestubbe</groupId>
<artifactId>FastCore</artifactId>
<version>0.1.0</version>
</dependency>
</dependencies>Add the repository and dependencies to your build.gradle:
repositories {
maven { url 'https://jitpack.io' }
}
dependencies {
implementation 'com.github.andrestubbe:FastAudioProcess:0.1.7'
implementation 'com.github.andrestubbe:FastSIMD:0.1.3'
implementation 'com.github.andrestubbe:FastCore:0.1.0'
}Download the latest JAR directly to add it to your classpath:
- REFERENCE.md: Exhaustive catalog of DSP algorithms, FFT kernels, timeline mixing, and acoustics architecture.
- CHANGELOG.md: Complete chronological record of version releases and performance improvements.
- PHILOSOPHY.md: Zero-allocation, native-first, and low-overhead real-time DSP design principles.
- COMPILE.md: Full compilation guide (MSVC Native AVX2 & Maven Build Setup).
- ROADMAP.md: Planned milestone features, ARM NEON SIMD, and GPU compute extensions.
| Platform | Architecture | Status | Vector ISA |
|---|---|---|---|
| Windows 10 / 11 | x64 (AMD64) | ✅ Fully Supported | AVX2 / FMA3 |
| Windows Server 2019+ | x64 | ✅ Fully Supported | AVX2 |
| Linux (Ubuntu/Debian) | x64 / ARM64 | 🚧 Planned | AVX2 / NEON |
| macOS (Apple Silicon) | ARM64 | 🚧 Planned | NEON |
- FastVoice — Zero-shot AI voice cloning and speech-to-speech conversion
- FastAudioCapture — Zero-copy WASAPI audio loopback & microphone capture
- FastAudioPlayer — Ultra low-latency WASAPI and DirectSound audio playback engine
- FastVAD — Ultra-fast real-time Voice Activity Detection (Silero-ONNX & WebRTC)
- FastTTS — Unified, zero-bloat text-to-speech backend orchestration
- FastSTT — High-throughput local speech-to-text recognition
- FastCore — Native library loader, FFM gateway, and platform abstraction
- FastSIMD — Hardware SIMD acceleration substrate
MIT License — See LICENSE for details.
Part of the FastJava Ecosystem.