How netease-youdao/emotivoice works
EmotiVoice is a multilingual, multi-voice text-to-speech engine with prompt-controlled emotion and style. Its main user path is text entry, phoneme conversion, prompt/speaker-conditioned synthesis, and audio output; users can access it through a web demo, batch scripts, or an OpenAI-compatible API. The repository also supports voice-cloning data preparation and model training. Arrows follow explicit README statements or source-index references; unsampled interface and training wiring is left unverified where necessary.
Main entry points: inference_am_vocoder_joint.py Batch inference · frontend.py Text frontend · model_open_source.py Prompt TTS model · jets.py JETS generator
Components 23
User interfaces
- Web demo
demo_page.py - OpenAI API
openaiapi.py - Batch inference
inference_am_vocoder_joint.py - HTTP API demo
HTTP_API_TtsDemo/apidemo/TtsDemo.py
Text processing
- Text frontend
frontend.py - Chinese frontend
frontend_cn.py - English frontend
frontend_en.py - Number conversion
cn2an/an2cn.py - Text cleaners
text/cleaners.py
Speech synthesis
- Synthesis config
config/joint/config.py - Style encoder
models/prompt_tts_modified/simbert.py - JETS generator
models/prompt_tts_modified/jets.py - Prompt TTS model
models/prompt_tts_modified/model_open_source.py - Text encoder
models/prompt_tts_modified/modules/encoder.py - Alignment module
models/prompt_tts_modified/modules/alignment.py - Variance module
models/prompt_tts_modified/modules/variance.py - HiFi-GAN vocoder
models/hifigan/get_vocoder.py - Generated audio
Data and training
- Voice data recipes
data - Alignment preparation
mfa - Acoustic features
models/prompt_tts_modified/feats.py - Model training
train_am_vocoder_joint.py
Other components
- TTS user
Connections 24
- TTS user → Web demo: enters text
README.md - TTS user → Batch inference: submits batch
README.md - TTS user → OpenAI API: sends request
README.md - TTS user → HTTP API demo: uses service demo
- Web demo → Text frontend: converts text
- Batch inference → Text frontend: prepares phonemes
README.md - Text frontend → Chinese frontend: dispatches Chinese
frontend.py - Text frontend → English frontend: dispatches English
frontend.py - Chinese frontend → Number conversion: converts numerals
frontend_cn.py - English frontend → Text cleaners: cleans text
- Batch inference → Style encoder: encodes prompts
inference_am_vocoder_joint.py - Batch inference → JETS generator: synthesizes speech
inference_am_vocoder_joint.py - Batch inference → Synthesis config: loads settings
inference_am_vocoder_joint.py - JETS generator → Prompt TTS model: uses model
models/prompt_tts_modified/jets.py - JETS generator → HiFi-GAN vocoder: uses generator
models/prompt_tts_modified/jets.py - Prompt TTS model → Text encoder: uses encoder
models/prompt_tts_modified/model_open_source.py - Prompt TTS model → Alignment module: uses alignment
models/prompt_tts_modified/model_open_source.py - Prompt TTS model → Variance module: uses variance
models/prompt_tts_modified/model_open_source.py - Batch inference → Generated audio: writes audio
inference_am_vocoder_joint.py - Voice data recipes → Text frontend: generates phonemes
data/DataBaker/src/step2_get_phoneme.py - Voice data recipes → Alignment preparation: prepares alignment
- Alignment preparation → Acoustic features: prepares training data
- Acoustic features → Model training: supports training
- OpenAI API → Batch inference: invokes synthesis