Missing sequences in transcription after Whisper update. #1193
Replies: 3 comments 1 reply
-
I have encountered something similar before. The cause was a delayed timestamp for the end of the previous window which caused it to start the next window too late and miss words. There may be different types of scenarios that can cause it, but one such scenario is addressed in #1114 when word timestamps are enabled. You can try it to see if it catches your case. |
Beta Was this translation helpful? Give feedback.
-
Bumping this one up. I frequently encounter this issue as well. |
Beta Was this translation helpful? Give feedback.
-
Makes an otherwise perfect transcriber pretty useless : (. medium.en on local here |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
After updating Whisper from the release
20230124
to20230314
, I noticed that thesmall.en
andlarge
models have issues with missing segments in transcriptions, mostly at the end or close to the end of the audio (for both transcriptions with and without word-level timestamps). As can be seen below transcription for the newer version is missing a segment between 03:00.000 and 03:18.100. What can cause it and is there any possible solution for this?Transcriptions example:
Source audio: Jeremy Renner to open up in exclusive interview with Diane Sawyer l GMA
Beta Was this translation helpful? Give feedback.
All reactions