Make subtitle lines shorter

[samuelbraun04]

How can one make the individual subtitle lines shorter?

Like instead of:

[0.00s -> 5.78s]  In the depths of history, peace treaties have often been celebrated as moments of triumph,
[6.10s -> 9.90s]  heralding the end of conflicts and the beginning of a new era of harmony.
[10.26s -> 14.52s]  Yet, not all peace treaties have lived up to these lofty expectations.

.transcribe() would return something like (this example is inaccurate but it has the right idea):

[0.00s -> 3.78s]  In the depths of history, peace treaties have 
[3.78s -> 5.50s]  often been celebrated as moments of triumph, 
[5.50s -> 7.65s]  heralding the end of conflicts and the beginning 
[7.65s -> 10.26s] of a new era of harmony. Yet, not all peace
[10.26s -> 14.52s]  treaties have lived up to these lofty expectations.

I've tried playing around with max_new_tokens and chunk_length but changing either of them just ruins the output. Is there another argument I'm missing?

[trungkienbkhn]

Unfortunately, FW does not support this.

[Purfview]

This repo doesn't have the subtitle writing functions, there are plenty of whisper implementations with own subs writers, after some research you should find a good subs writing function.

For heads-up, the reference Whisper's subs writers are wonky...

On my repo I implemented good splitting capabilities, but it's not for Python.
Another good alternative is to use Subtitle Edit, but splitting there will not be perfect because it wouldn't have access to the word timestamps.