voicenote

Voice-powered note taking with whispercpp & ChatGPT.

This script couples well with a Markdown renderer that supports auto-refresh, here's what I use.

Usage

• Install the following libraries (pip install ...):
  • sounddevice, numpy, whispercpp, openai
• You will need an OpenAI API key (https://platform.openai.com/account/api-keys).
• Setup the API key enviornmental variable: export OPENAI_API_KEY='sk-...'
• Just run python3 main.py.

Todo

• Handle context better
  • Right now the request is in this format: [system prompt, latest file version, transcription]
  • Occasionally we should get a summary of the notetaking so far and add it as a system message or something to cut down on tokens
• Add GUI
• This might be a nice Obsidian plugin
• Audio recording and handling of messages could be done better