A visual speech recognition (VSR) tool that reads your lips in real-time and types whatever you silently mouth. Runs fully locally.

Relies on a model trained on the Lip Reading Sentences 3 dataset as part of the Auto-AVSR project.

Watch a demo of Chaplin here.

1. Clone the repository, and cd into it:

   git clone https://github.com/amanvirparhar/chaplin
   cd chaplin

2. Download the required model components: LRS3_V_WER19.1 and lm_en_subword.
3. Unzip both folders, and place them in their respective directories:

   chaplin/
   ├── benchmarks/
       ├── LRS3/
           ├── language_models/
               ├── lm_en_subword/
           ├── models/
               ├── LRS3_V_WER19.1/
   ├── ...
   

4. Install and run ollama, and pull the llama3.2 model.
5. Install uv.

1. Run the following command:

   sudo uv run --with-requirements requirements.txt --python 3.12 main.py config_filename=./configs/LRS3_V_WER19.1.ini detector=mediapipe

2. Once the camera feed is displayed, you can start "recording" by pressing the option key (Mac) or the alt key (Windows/Linux), and start mouthing words.
3. To stop recording, press the option key (Mac) or the alt key (Windows/Linux) again. You should see some text being typed out wherever your cursor is.
4. To exit gracefully, focus on the window displaying the camera feed and press q.