shivammehta25/Matcha-TTS

Based on the README, here's the technical summary: Implements conditional flow matching with ODE-based synthesis to achieve fast non-autoregressive TTS, supporting variable inference steps via Euler solver and controllable generation through temperature and speaking rate parameters. Built on PyTorch Lightning with Hydra configuration, it includes ONNX export/inference support and a Gradio interface for browser-based synthesis. Enables phoneme-level alignment extraction and supports multi-GPU training on custom datasets with automatic mel-spectrogram normalization.

1,259 stars.