• Missing voice recordings...
  • What
    • Music bot
    • Justin's audio part of the time... why?
  • What to do
    • Review voice.rs changes - diff_from_last_timestamp was crashing if < 0; I did a quick fix during game to stop crashing but needs review. May have led to inaccurate timestamps or missed recordings?
    • Potentially need to change "tick that looks like continuous voice from before" flushing
    • known_ssrcs - voice data that comes in from a NOT known ssrc is skipped. Instead, let's still record it - just change the filename if necessary
      • Log data about this
• Logs too noisy / not useful enough
  • Separate log for levels - i.e. verbose.log for VoiceTick that's repeated thousands of time; regular log for more important data
• Keep track of who's talking - real time?
  • Would be awesome to have a little websockets page with realtime data about who's talking. We already get packet updates on server every 20ms - perfect amount of time to not deal w/ timing myself. Update state with list of all current speakers and display in real time preview
  • Could add other useful data on this page too
  • Also add inputs to bot - join channel, send message...
• Have bot join voice channel without me needing to send public message
  • Need a way to trigger this - from a REST or websocket listener
  • And/or auto join specific channel on startup
    • CLI parameter - channel id?

Models (OpenAI): gpt-3.5-turbo, whisper-1

• Messaging
  • Reply detection
  • Rate limiting
• Comprehensive logging
• Music
  • YouTube search
  • Queue controls
• Voice
  • Live transcriptions
  • Transcription-based replies
  • Text to speech
  • Music controls

For local development:

• rust
• ffmpeg
• opus
• yt-dlp

OR

For Dockerized development:

• docker
• docker-compose

Create a .env file from .env.example, then tweak src/cfg.rs to your needs.

Running:

# Locally
cargo run

# Using Docker
docker-compose up

• python@^3.11
• poetry

Create a new file model/<name>.jsonl and update the path in model/tune.py. Alternatively, update model/train.jsonl directly.

To queue up a fine-tuning job on OpenAI:

cd model
poetry shell
poetry install
poetry run python tune.py

License