Timing of subtitles is way off if I limit max_line_count and max_line_width==bad things?

[adhocinterest]

This is the command I used

whisperx ~/Downloads/Telegram\ Desktop/purpose\ of\ govt.mp3 --max_line_count 7 --max_line_width 11 --model small 
--batch_size 1 --compute_type float32

The transcription timing is way off. The first couple sections have the exact same start and end and the the next few have text in areas where there is silence or the words are not part of that section.

I can supply the audio if necessary. Is this part of the bad things can happen because of the float32 due to my hardware? Or is there a deeper issue?

[adhocinterest]

As a side note, I can't tell for sure yet, but the JSON file seems to be accurate. I've got to put it in my editor to validate that's the case, but at first glance it definitely doesn't have the problems the set and vtt files have.

[nikola1975]

I did not notice a problem in the scope you are mentioning. Have you confirmed the problem with various files?

From my experience, WhisperX subtitles are more precise than OpenAI Whisper's API implementation.

[adhocinterest]

I just tried it with a different file and it didn't seem repeat, but it did repeatedly on the previous file. My purpose has changed somewhat and the file format will not fit my needs any longer. I instead will be creating what I need by parsing the json into what I need. Thank you. I'll go ahead and close the issue with this comment as I don't plan on assisting in troubleshooting anymore.