Begin investigating STT for audio-only code generation

[j6k4m8]

https://github.com/coqui-ai/STT

[j6k4m8]

Looks like whisper might also have faster-than-realtime performance for some of the smaller models; could consider running our own STT server for high-speed or realtime streaming responses.