I made this demo in 2-3 hours, and I did use the "wait until the dictation results are finalized" technique which is safer (i.e. the dictation transcription is more robust) but slower.
For another demo - https://www.youtube.com/watch?v=fxS7OKh_4vc - I kept feeding the "in progress" transcription results into GPT and that was super super awesome & fast. It would just require more work to deal with all of the different timings going on (i.e. there's the speech itself from the person, the time to transcribe, sending the request to GPT, "sync'ing" it to where the person is (mentally/in their speech) at the point where GPT replies, etc.)
But yeah. Real time/continuous talk is absolutely where it's at. Should GPT be available as a websocket...?!
I made this demo in 2-3 hours, and I did use the "wait until the dictation results are finalized" technique which is safer (i.e. the dictation transcription is more robust) but slower.
For another demo - https://www.youtube.com/watch?v=fxS7OKh_4vc - I kept feeding the "in progress" transcription results into GPT and that was super super awesome & fast. It would just require more work to deal with all of the different timings going on (i.e. there's the speech itself from the person, the time to transcribe, sending the request to GPT, "sync'ing" it to where the person is (mentally/in their speech) at the point where GPT replies, etc.)
But yeah. Real time/continuous talk is absolutely where it's at. Should GPT be available as a websocket...?!