The OpenAI Realtime API is annoyingly good.

Yappatron’s implementation of it is completely perfect and real-time. The transcription finally just behaves. There is honestly nothing else like this.

The implementation is very straightforward. Yappatron opens a transcription WebSocket, streams microphone audio to gpt-realtime-whisper, and writes the transcript into whatever I am using. You plug it into a thing, and then there you go.

I just want to have this on 24/7.

The frustrating part is that the API is expensive.

But the API is very worth it.