SenseVoiceDemo

This is a demo of SenseVoice.

The model.py is copied from SenseVoice.

How to run

1. Make sure you have installed conda.

2. Run conda env create -f environment.yml to create the environment.

3. Run conda activate SenseVoiceTest to activate the environment.

4. Download the pre-trained model from modelscope or huggingface.

   For example, download the model from modelscope and put it in the model folder:

   • Run git lfs install to install git-lfs;
   • Run cd iic to enter the iic folder;
   • Run git clone https://www.modelscope.cn/iic/sensevoicesmall.git to clone the model.
   • At the end, folder structure should be like this:

   SenseVoiceDemo
   |   .gitignore
   |   environment.yml
   |   LICENSE
   |   main.py
   |   model.py
   |   README.md
   |   requirements.txt
   |
   +---iic
   |   \---SenseVoiceSmall
   |       |   .gitattributes
   |       |   am.mvn
   |       |   chn_jpn_yue_eng_ko_spectok.bpe.model
   |       |   config.yaml
   |       |   configuration.json
   |       |   model.pt
   |       |   README.md
   |       |
   |       +---example
   |       |       .DS_Store
   |       |       en.mp3
   |       |       ja.mp3
   |       |       ko.mp3
   |       |       yue.mp3
   |       |       zh.mp3
   |       |
   |       \---fig
   |               aed_figure.png
   |               asr_results.png
   |               inference.png
   |               sensevoice.png
   |               ser_figure.png
   |               ser_table.png
   

5. Run pip install -r requirements.txt to install the required packages.

6. Run python main.py to run the demo.

7. By the way, If you have an NVIDIA GPU, you can modify the device in main.py to cuda:0 to use GPU:

    model = AutoModel(
        model=model_dir,
        trust_remote_code=True,
        remote_code="./model.py",  
        vad_model="fsmn-vad",
        vad_kwargs={"max_single_segment_time": 30000},
        device="cuda:0",
        # device="cpu",
    )