Whisper AI issues when transcribing longer files

Support & Feedback Forum
,
1

I’ve been importing longer podcasts (20+ minutes) into LingQ using Whisper and I noticed that it often has ‘dead zones’. The transcribed text will cut out for minutes, mostly in the middle of the episode. Often there’s multiple dead zones. I think so far I have not had a single podcast episode without them. I even tried breaking up the episodes into shorter audio clips of 15minutes but it did not help.

For an example I can give this episode Login - LingQ

It has multiple dead zones. The first deadzone is when the transcription text is about halfway, the audio is only around 33% in. Another is the ending. After the transcription text ends the audio still has ~10 minutes left.

It’s an extremely useful feature and I use it a lot. Does anyone have any tips in how to effectively transcribe the audio? Is there a limit on the clip length that works best?

3 Likes
2

Thanjs for reporting this. I asked our team to investigate.

1 Like
3

I have had these issues even with shorter audios. It seems like Whisper got more sensitive and even small interferences do interrupt the transcription. I had this issue when music snippets appeared in the podcast, but recently it has been hard to even find a reason why Whisper could have struggled with the transcription.

1 Like