Audio Transcriber is a tool that allows users to upload audio files via a custom API platform. The backend processes the audio files, converting them to WAV format, and then performs a transcription in text form. Users can select different language models for precise transcriptions, and the transcribed text is further processed by a language model (e.g., ChatGPT) to generate an accurate summary. Both the transcribed text and the summary are provided via the API.
- Development of a custom API platform
- Users can upload audio files via the API
- The backend receives the audio files and converts them to WAV format
- Subsequently, a transcription is performed in text form
- Integration of speech recognition and/or language selection for different languages
- Option to select different language models (e.g., Whisper) for precise transcriptions
- The transcribed text is sent to a language model (e.g., ChatGPT) to generate an accurate summary
- The transcribed text and summary are provided via the API.
Planned features and enhancements can be found on the project's board. Check out the board for updates and future developments.
- Ensure Python 3.8 or later and NodeJS is installed.
- Clone the repository.
- Navigate in the console to the root folder of the repository.
- OPTIONAL: Create a virtual environment with
python -m venv .venv
if needed. Otherwise, proceed to step 6. - OPTIONAL: Activate the virtual environment.
- Install all required Python packages by running
pip install -r requirements.txt
. - Install PyTorch from here.
- OPTIONAL: When using a supported OS, install Accelerate to benefit from performance gains. Check here for instructions.
- OPTIONAL: Install Deepspeed to benefit from additional performance gains. Instructions can be found here.
- Rename
config.example.py
toconfig.py
and configure to your liking. - Start the backend by executing
main.py
in the backend folder. - Start the frontend by typing
ng serve
into the console inside the frontend folder.
For more information on how to contribute, please visit the Contributing page.