I'm using Premiere Pro v 22.4.0 (Build 57) on an iMac (Retina 5k) and tried out the auto transcription service on an interview sequence (using a mix of 4 clear channels of audio). The transcription itself is radically improved from the last time I used it, however the timecode functionality seems WAY off and also counts backwards -- it appears to have little to do with the sequence timecode at all.
In my sequence (about 57 minutes, clips from 00:00:00:00 to 00:57:20:12), the interview clip begins at 00:00:00:00, but in the transcript it begins at 08:00:00:03.
The first spoken sentence goes from 00:00:00:00 to around 00:00:17:16, but in the transcript, this is 08:00:00:03 to 09:59:43:06.
After that, the timecode begins counting backwards and has no relationship to the sequence. e.g. Second spoken line is 09:59:42:17 - 09:59:26:09.
This continues on down the transcript to the last spoken sentence, which in the sequence runs from 00:57:06:17 to 00:57:20:12, but in the transcript 09:02:51:23 - 09:02:43:02.
(Autoscroll is also, hence, broken. As I play back the sequence, it flies to the bottom of the transcript.)
This is confounding, because from a computing perspective, it would seem that the timecode is the easier computing task than the voice to text technology. My entire purpose of generating the transcripts is to have a reference to be able to skim/highlight and return to the timecode easily.
I tried re-transcribing, but this time selecting only one track. The results were exactly the same except + 1 frame of timecode everywhere.
The exported timecodes are the same as what appears in Premiere.
Is anyone else having very strange timecode outcomes with the auto transcribe feature?