This project allows you to locally run an application that will transcribe your conversational data with privacy and security. Transcripts are generated and stay on your device. Depending on your machine specification, you can use the base, medium, or large whisper models.
This does not currently add diarization, but that can be backfilled on the full transcript using tools like whisper-diarization.
Outputs will be stored to a generated folder with a name like transcriptions_yyyymmdd_timestamp.
The script depends on pyaudio, so you may also need to install portaudio on your machine.
brew install portaudio # MacOS
python -m pip install pyaudio # Windows
sudo apt-get install python3-pyaudio # LinuxOnce you have that, run the live_transcribe script to get going.
git clone [email protected]:sabaimran/vox-locus.git && cd vox-locus
uv venv --python 3.12
source .venv/bin/activate
uv pip install -r pyproject.toml
python3 live_transcribe.pyRun Ctrl+C to kill the recording and get your final transcripts.
There is also a record.py script that just takes in your audio and dumps it into a .wav file for any post-processing you might want to do.