You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've been playing with this repo some using the faster-whisper backend and the large-v3-turbo model, and most of the time it works really well However, there are some cases where an incomplete segment will get longer and longer, and then move on to the next segment without that long segment (24 seconds) never getting marked as complete.
Why is this? Is it because there wasn't a long enough silence or something between the segments? How can I make sure that a long segment with a high confidence (as determined by avg_logprob) gets marked as complete before moving on?
I was going to just mark it as complete before moving to the next segment, but the next segment overlaps, so there'd be overlapping text returned to the user.
(Not sure if it's relevant, but I do have two clients connected at the same time)
The text was updated successfully, but these errors were encountered:
I've been playing with this repo some using the faster-whisper backend and the large-v3-turbo model, and most of the time it works really well However, there are some cases where an incomplete segment will get longer and longer, and then move on to the next segment without that long segment (24 seconds) never getting marked as complete.
Why is this? Is it because there wasn't a long enough silence or something between the segments? How can I make sure that a long segment with a high confidence (as determined by avg_logprob) gets marked as complete before moving on?
I was going to just mark it as complete before moving to the next segment, but the next segment overlaps, so there'd be overlapping text returned to the user.
(Not sure if it's relevant, but I do have two clients connected at the same time)
The text was updated successfully, but these errors were encountered: