Improved transcription
right now the transcription got an upgrade so it's much more accurate, but cold start takes about 45 seconds. So, having a user just transcribe one cell and then move on really doesn't make sense unless we put it on an always warm GPU, which would be a different setup than what we have on modal.com. In any case, it might be best to require users to transcribe more than one thing at a time. Otherwise, it's just not worth it.
Also, we need to pass the language code to the model because it now takes that and that should greatly improve transcription quality.
Improved transcription
right now the transcription got an upgrade so it's much more accurate, but cold start takes about 45 seconds. So, having a user just transcribe one cell and then move on really doesn't make sense unless we put it on an always warm GPU, which would be a different setup than what we have on modal.com. In any case, it might be best to require users to transcribe more than one thing at a time. Otherwise, it's just not worth it.
Also, we need to pass the language code to the model because it now takes that and that should greatly improve transcription quality.