Most of my life is based around knowledge work, and the processes of thinking. Most of my best thinking isn't done at my desk, so I frequently talk through problems in my own head while walking, or doing other physical tasks away from a screen.
This project was an experiment at capturing some of my thoughts by recording things I've talked to myself about in an audio file, and distilling the key points.
Project is built on Python 3.11, and utilizes the OpenAI API's to access the Whisper (Audio Transcription) and GPT-4 (LLM) models.
The primarily limit of the app in it's current form is the 16k tokens used in the GPT-4 model (available in private beta). It cannot handle large transcripts from long-running recordings. My average speech rate when talking to myself is about 80WPM (weird fact, I know). Assuming a conversion of words:tokens of 0.66, that gives at least two hours of conversation in a single run.
- Clone the repository.
- Setup the necessary modules using
pip. - Run the script using
./python r3.py recording.mp3.