Vision: To provide the fastest possible native and SIMD audio processing primitives for the FastJava ecosystem, optimizing where standard Java APIs fail.
- Format Conversions: MP3 to WAV conversion.
- Resampling: Wave resampling to 44100Hz.
- SIMD RMS Volume Calculation: High-performance volume level computation accelerated by Java 17 Vector API (SIMD).
- Wake-Word Integration Utilities:
- Frame Normalization: Energy-sensitive amplitude normalization.
- Band-limited Pre-Emphasis Filter: High-pass filtering to boost higher speech frequencies.
- Modellfriendly Frame-Chunking Pipelines: Buffering audio frames tailored to wake-word neural networks.
- Log-Mel Feature Extraction: Ultra-fast extraction of MFCCs / Log-Mel spectrograms for local speech models.
- Audio-Manipulation & FX Library:
- Equalizer (IIR/FIR crossover filters)
- Dynamic Processing (Noise Gate)
- Pitch-Shifting (Native speed-preserved & resampling-based)
- Audio I/O & Mixing:
- Multi-channel mixing (downmix / upmix)
- Waveform Visualization & Metering:
- In-memory signal downsampling (
generateWaveformPoints) for timeline plotting. - Absolute peak amplitude tracking (
getFramePeak) for real-time streams.
- In-memory signal downsampling (
- Whisper & TTS Substrates:
- Native integration helper for Whisper (STT) and local TTS engine (Style-TTS/VITS).
- VAD Trigger Integration:
- Stable Voice Activity Detection (VAD) runtime bindings.
- Fast Audio Embedding extraction:
- CLAP/Wav2Vec ONNX extraction.
- AI Noise Reduction:
- RNNoise / Demucs JNI bindings.
- CREAM Integration:
- Audio event markers on timetables and timeline node snapshots.
Focus: Performance is our USP. We optimize where Java stops.