metadata
license: mit
This is a streamlined interface version of WavTokenizer-large-speech-75token, providing a clean, efficient way to interact with the model through separate encoder and decoder components.
- Reduced model size from 1.75GB to ~330MB by keeping only necessary components for inference
- Split interface (82MB encoder, 248MB decoder)
- Simplified integration with just one .py file
The model is split into:
encoder/: Handles audio encodingdecoder/: Handles decoding and synthesis