v0.4.0 - Streaming transcriptions 🎙️

What's new

Streaming transcription support: you can now get the transcript as it's being decoded instead of waiting for the whole thing.

Send stream=true to /v1/audio/transcriptions and you'll receive transcript.text.delta events as text comes out, followed by a final transcript.text.done with the complete result. This is OpenAI's streaming-transcription protocol, so existing OpenAI-compatible clients work out of the box.

How to use it

curl -N -X POST http://localhost:5092/v1/audio/transcriptions \
  -F "file=@speech.wav" \
  -F "response_format=json" \
  -F "stream=true"

You'll see:

event: transcript.text.delta
data: {"type":"transcript.text.delta","delta":"Maybe"}

event: transcript.text.done
data: {"type":"transcript.text.done","text":"Maybe next time, huh?"}

Notes

For live microphone input (transcribe-while-you-speak), you'd want OpenAI's Realtime/WebSocket API: that's a different protocol and not part of this release.