AudioSig
AudioSig is a portable, NumPy-only signal-processing layer for speech and TTS applications. It provides waveform effects, spectral and pitch analysis, and reference metrics without model downloads or neural inference. STOI/ESTOI require explicit equal-length mono input at 10 kHz.
Public behavior
The package accepts float32 and float64 NumPy arrays and keeps the
sample axis explicit through the public APIs. split returns clipped,
half-open non-silent sample intervals. normalized_energy_vad preserves its
historical non-padded default; pass pad_end=True when trailing partial
speech must be analyzed. find_speech_bounds returns [0, 0] when no
speech is found.
See the repository README for the detailed replacement boundary and downstream audit status.
Documentation
Contents:
- Getting Started with AudioSig
- AudioSig API Reference
- AudioSig Examples
- Advanced Topics
- Choosing a time-stretch backend
- Time-varying speech envelopes
- Resampling Internals
- Loudness measurement
- VAD Algorithm Details
- Edge Cases and Special Handling
- Multi-Channel and Batch Processing
- Error Handling Strategies
- Performance Optimization
- Integration with Other Libraries
- Debugging and Profiling
- Speech effects quality evaluation
- Changelog