How rhasspy/piper works
Piper provides neural text-to-speech inference and tooling to prepare data, train voice models, and export them for runtime use. A user can submit text through the Python CLI/server or C++ executable, select or download a voice, and synthesize WAV or streamed audio. Training covers dataset preprocessing, VITS modeling, inference, and export; voice conversion and speaker/audio utilities are additional capabilities. Source excerpts and the index confirm some connections; dashed, uncited links mark important training or export relationships not verified in the sampled source.
Main entry points: voice.py Python Voice · piper.cpp C++ Synthesis · lightning.py VITS Training · __main__.py Python CLI
Components 24
Python Runtime
- Python CLI
src/python_run/piper/__main__.py - HTTP Server
src/python_run/piper/http_server.py - Voice Download
src/python_run/piper/download.py - Python Voice
src/python_run/piper/voice.py
C++ Runtime
- C++ CLI
src/cpp/main.cpp - C++ Synthesis
src/cpp/piper.cpp
Training Workflow
- Dataset Preprocess
src/python/piper_train/preprocess.py - Audio Normalize
src/python/piper_train/norm_audio/__init__.py - Training Commands
src/python/piper_train/__main__.py - VITS Training
src/python/piper_train/vits/lightning.py - Training Inference
src/python/piper_train/infer.py - Model Export
src/python/piper_train/export_onnx.py - Voice Conversion
src/python/piper_train/voice_conversion.py
VITS Engine
- VITS Models
src/python/piper_train/vits/models.py - VITS Layers
src/python/piper_train/vits/modules.py - VITS Operations
src/python/piper_train/vits/commons.py - VITS Transforms
src/python/piper_train/vits/transforms.py - Mel Processing
src/python/piper_train/vits/mel_processing.py
Other components
- Text User
- ONNX Runtime
- Voice Files
- Phonemization
- WAV Audio
- Audio Dataset
Connections 28
- Text User → Python CLI: submits text
src/python_run/piper/__main__.py - Text User → HTTP Server: sends request
- Python CLI → Voice Download: resolves voice
src/python_run/piper/__main__.py - HTTP Server → Voice Download: resolves voice
src/python_run/piper/http_server.py - Python CLI → Python Voice: loads voice
- HTTP Server → Python Voice: uses voice
src/python_run/piper/http_server.py - Voice Download → Voice Files: downloads files
src/python_run/piper/download.py - Python Voice → ONNX Runtime: runs inference
src/python_run/piper/voice.py - Python Voice → Phonemization: phonemizes text
src/python_run/piper/voice.py - Python Voice → WAV Audio: writes WAV
src/python_run/piper/voice.py - Text User → C++ CLI: provides text
src/cpp/main.cpp - C++ CLI → C++ Synthesis: synthesizes text
src/cpp/main.cpp - C++ Synthesis → Phonemization: phonemizes text
src/cpp/piper.cpp - C++ Synthesis → ONNX Runtime: runs inference
src/cpp/piper.cpp - C++ Synthesis → WAV Audio: writes audio
- Audio Dataset → Dataset Preprocess: provides recordings
- Dataset Preprocess → Audio Normalize: normalizes audio
src/python/piper_train/preprocess.py - Dataset Preprocess → Training Commands: prepares training data
- Training Commands → VITS Training: starts training
- VITS Training → VITS Models: trains model
- VITS Models → VITS Layers: uses layers
- VITS Layers → VITS Operations: uses operations
src/python/piper_train/vits/modules.py - VITS Layers → VITS Transforms: uses transforms
src/python/piper_train/vits/modules.py - Audio Normalize → Mel Processing: computes spectrogram
src/python/piper_train/norm_audio/__init__.py - VITS Training → Training Inference: supports inference
- VITS Training → Model Export: exports trained model
- Model Export → Voice Files: produces voice model
- Training Inference → Voice Conversion: supports conversion