retranscribe with feedback
Closed this issue · 5 comments
transcription quality of even bad audio files is surprisingly good; however, some words are wrongly transcribed throughout the audio (like names of places etc.).
it would be great to have a way to make the engine learn on user feedback, maybe through the assistant (like chat-gpt is also able to learn on user feedback).
You can force attention on names etc. using the Initial Prompt field - just add them there.
just add the name of the place? doesn't seem to work in my case, see this (portuguese) example:
Sou mesmo junto ao Rock in Rio, de Shellas, de Lisboa.
Ah, de Shellas.
"Shellas" is a good phonetical transcription. The place's name is "Chelas", however, and it appears many times during the audio.
adding an instruction like "Shelas should be written Chelas" to the Initial Prompt field doesn't do the job either. I guess I don't really understand how the Initial Prompt field works.
Just noticed that you replied on this, sorry!
Sou mesmo junto ao Rock in Rio, de Shellas, de Lisboa.
Ah, de Shellas.
Try to actually write the corrected version of that in the Initial Prompt field:
- Sou mesmo junto ao Rock in Rio, de Chelas, de Lisboa.
- Ah, de Chelas.
More so, you can even try to re-transcribe only the portions of the text that contain that particular name.
But, if the thing only occurs 2-3 times in the transcription, I recommend simply editing the transcription itself of course.
Cheers!
That worked, thanks a lot!