Westlake-Omni

Westlake-Omni: Open-Source Chinese Emotional Speech Interaction Large Language Model with Unified Discrete Sequence Modeling

🤗 Hugging Face | 📖 Github

Westlake-Omni is an open-source Chinese emotional speech interaction large language model that utilizes discrete representations to achieve unified processing of speech and text modalities. The model supports low-latency generation and high-quality Chinese emotional speech interaction.

Highlights

🎙️ Utilizes discrete representations to unify the processing of speech and text modalities..

🎭 Trained on a high-quality Chinese emotional speech dataset, enabling native emotional speech interaction in Chinese.

⚡ Low-latency speech interaction, simultaneously generating text and speech responses.

https://github.com/user-attachments/assets/02a71c01-3384-4845-8e7f-4e0dda35d8f3

Install

Create a new conda environment and install the required packages:

conda install pytorch==2.3.0 torchvision==0.18.0 torchaudio==2.3.0 -c pytorch

git clone [email protected]:xinchen-ai/Westlake-Omni.git
cd Westlake-Omni
pip install -r requirements.txt

Quick start

Interactive demo

run gradio demo

python gradio_demo.py

Local test

python generate.py --user-audio data/sounds/input.wav --user-text 嗯，最近心情不是很好，能聊聊吗？
python generate.py --user-audio data/sounds/input.wav

Acknowledgements

fish-speech The codebase we built upon.
Qwen2 as the LLM backbone.

Lincese

The current code and the vqgan model weights are provided under the CC-BY-NC-SA-4.0 License. The large language model weights are provided under the Apache 2.0 License. Note that parts of this code are based on Fish speech, released under the CC-BY-NC-SA-4.0 License.

Contact

If you have any questions, please raise an issue or contact us at [email protected].

xinchen-ai
/

Westlake-Omni