- Download and compile whisper.cpp (follow steps there to download model weights as well)
- Clone this repo
- Add a
.env
file containing the following keys:CORTEX_DOWNLOAD_PATH
, set to the full path of theepisodes/
directory;WHISPER_PATH
, set to the full path of the directory containing the whisper binary.
- Download and transcribe episodes:
conda env create -f environment.yml
conda activate crtx
# Download episodes
python download.py
# Transcode episodes from .mp3 to .wav using ffmpeg
chmod +x encode.sh
./encode.sh
# Transcribe downloaded episodes
python transcribe.py
This will create a list of csv
files containing whisper transcriptions of each episode in the form start_seconds, end_seconds, transcription
.