-
|
If audio file is long (more than 20min), suddenly whisper makes too many timestamps from approximately 20min and phrasess are divided into words like this: [00:19:34] I met a very nice friend in a college. From the beginning, whisper doesn't splits phrases, but from some moment, starts doing so. I checked if the word level timestamp option is on, but it was turned off. Are there any ways to fix this? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment
-
|
There's a certain level of randomness in Whisper's behavior in splitting segments, and unfortunately it's path dependent so that when it starts to happen, it tends to continue happening. In the case of excessively splitting phrases like the above, an easy fix is to post-process the transcript by merging lines that are too short .. |
Beta Was this translation helpful? Give feedback.
There's a certain level of randomness in Whisper's behavior in splitting segments, and unfortunately it's path dependent so that when it starts to happen, it tends to continue happening. In the case of excessively splitting phrases like the above, an easy fix is to post-process the transcript by merging lines that are too short ..