incorrect timestamps

[sloganking]

Output transcriptions such as .vtt and .srt files, have correct transcriptions, but the timestamps for what is said when is wrong. A 3 second sentence is marked as 3 minutes instead. It doesn't appear to be exact seconds to minutes though.

What we're outputting

00:00.000 --> 03:36.000
- Things you prioritize, like the most useful
03:36.000 --> 07:44.000
and interesting conversations, it goes parties,
07:44.000 --> 10:52.000
then workshops, then conference session.
10:52.000 --> 15:28.000
- I'm sorry, but you asked to be roasted, I will roast.
15:28.000 --> 17:56.000
- I'm talking about the level of intelligence
17:56.000 --> 20:52.000
of the cat or dog, okay?

What is outputted by yt-whisper and should be output instead

00:00.000 --> 00:06.500
Things you prioritize, like the most useful and interesting conversations, it goes parties, then workshops, then conference sessions.

00:06.500 --> 00:09.200
I'm sorry, but you asked to be roasted, I will roast.

00:09.200 --> 00:12.500
I'm talking about the level of intelligence of a cat or a dog, okay?

[m1guelpf]

@sloganking Could you check if 3f7558a fixes this?

[sloganking]

@m1guelpf That did not fix is. A 4 second sentence is still registered as 10 minutes long.

[sloganking]

Printing out fragment.start and fragment.stop in as_vtt(). They don't seem to correlate with milliseconds at all.

fragment: Utternace { start: 0, stop: 624, text: " Hello world, this is my second test." }
fragment: Utternace { start: 624, stop: 696, text: " Can you hear me very well?" }
fragment: Utternace { start: 696, stop: 1696, text: " [BLANK_AUDIO]" }

This is the output from a 8.3 second audio clip. The first sentence is finished in just under 4 seconds. Yet I don't see how 624 relates to 4s at all.