Segment order regression since 10mb chunking

[iandundas]

Since https://github.com/argmaxinc/WhisperKit/pull/158 was merged, we're seeing segments being delivered in the wrong order, including in the example app.

8fcfadbe37 (correct before)	25a07498 (incorrect after)

Settings:

Sample file: http://172.104.253.215/atp-7-min-clip.m4a

Full transcripts:

Full correct transcript Full incorrect transcript

[ZachNagengast]

Looking at this shortly, do you have any sense of what parts specifically changed between then? Might give a clue

[iandundas]

I don't have a great handle on it, it seems completely reordered and some segments are missing

For example, in the correct transcription the word "easter" occurs once:

[WhisperKit] [Segment 115] [474.04 --> 476.70] So, you know Easter just happened in.

Whilst in the bad transcription it appears four times:

meanwhile, the first line of the good transcription contains

[WhisperKit] [Segment 0] [0.00 --> 30.00] Do you have also just finishing listening to the hot pockets episode?

whilst this doesn't appear in the bad transcription at all

[iandundas]

good.txt bad.txt