How iszhanjiawei/indextts2 works
IndexTTS2 is a zero-shot expressive text-to-speech system. A user supplies text and a speaker-reference recording, optionally adding an emotion reference, emotion scores, or text emotion guidance; inference generates speech with either controlled or freely generated duration. The README documents these modes, and the source index connects the inference entry point to the main model and audio-generation modules. Several internal synthesis stages are represented by their implementation directories because their source was not sampled; those relationships are not inferred beyond documented capabilities and indexed references.
Main entry points: infer_indextts2.py Inference API · model_v2.py Autoregressive model · webui.py Web interface · bigvgan.py BigVGAN vocoder
Components 15
User access
- Web interface
webui.py - Inference API
indextts/infer_indextts2.py
Prompt conditioning
- Text frontend
indextts/utils/front.py - Audio features
indextts/utils/feature_extractors.py - Speaker encoding
indextts/s2mel/modules/campplus/DTDNN.py - Emotion guidance
indextts/infer_indextts2.py
Speech generation
- Autoregressive model
indextts/gpt/model_v2.py - Conformer encoder
indextts/gpt/conformer_encoder.py - Perceiver
indextts/gpt/perceiver.py - Duration adaptation
indextts/gpt/model_v2.py
Audio synthesis
- Mel generation
indextts/s2mel/modules - Codec components
indextts/utils/maskgct/models/codec - BigVGAN vocoder
indextts/s2mel/modules/bigvgan/bigvgan.py
Other components
- TTS user
- Generated audio
Connections 16
- TTS user → Web interface: submits text
README.md - TTS user → Inference API: provides prompts
README.md - Web interface → Inference API: invokes inference
webui.py - Inference API → Text frontend: uses frontend
indextts/infer_indextts2.py - Inference API → Audio features: uses extractors
indextts/infer_indextts2.py - Inference API → Speaker encoding: uses encoder
indextts/infer_indextts2.py - Inference API → Emotion guidance: accepts guidance
README.md - Inference API → Autoregressive model: uses model
indextts/infer_indextts2.py - Autoregressive model → Conformer encoder: uses encoder
indextts/gpt/model_v2.py - Autoregressive model → Perceiver: uses perceiver
indextts/gpt/model_v2.py - Autoregressive model → Duration adaptation: controls duration
README.md - Inference API → Mel generation: uses synthesis modules
indextts/infer_indextts2.py - Inference API → Codec components: uses codec utilities
indextts/infer_indextts2.py - Inference API → BigVGAN vocoder: uses vocoder
indextts/infer_indextts2.py - Inference API → Generated audio: writes speech
README.md - BigVGAN vocoder → Generated audio: synthesizes waveform