# Notice

This repository is built from BayLing-Duplex and uses upstream GLM-4-Voice components.

The model uses upstream components including:

- the upstream checkpoint format and tokenizer files;
- the upstream speech tokenizer;
- the upstream speech decoder based on CosyVoice components.

Code under `bayling_duplex_runtime/cosyvoice` contains files derived from CosyVoice and related components with Apache-2.0 headers. Keep the original source headers when redistributing or modifying those files.

For upstream GLM-4-Voice information, see:

- https://github.com/THUDM/GLM-4-Voice
- https://github.com/THUDM/GLM-4