Truncating Realtime Audio Transcriptions

API
1

Does anyone know of a way to figure out where to truncate transcriptions if you interrupt during response for the realtime API?

I have the API working and I can truncate the audio response. It is clear that the model knows that I cut it off and roughly where, but I am left with the transcript as far as it generated. If I were to restore the conversation in another session, I’d be entering the transcripts as far as they were generated which could be significantly ahead of where the audio got cut off and alter the context significantly.

2 Likes
2

I also have the same problem. I am thinking of storing the audio locally and then resending the truncated audio to whisper manually to get the updated text. But be great if I didn’t have to!

1 Like
3

Yeah I think this is the way

1 Like