Need help because I am obtaining TEXT LINES that are too long that cause next lines to be very SMALL sometimes

[BestofthebestinAI]

Example of Txt or Srt:

Txt:

Line 1: So like like termites don't even break it down and they have that protozoa in their gut that eat the the fiber and and then produce short-chain fatty acids which are 100% saturated and then those protozoa die off and the the bug absorbs them as protein the cows and and horses and
Line 2: Gorillas, etc.

in SRT:

489
00:37:34,441 --> 00:37:55,324
So like like termites don't even break it down and they have that protozoa in their gut that eat the the fiber and and then produce short-chain fatty acids which are 100% saturated and then those protozoa die off and the the bug absorbs them as protein the cows and and horses and

490
00:37:55,304 --> 00:37:56,485
Gorillas, etc.

How can I avoid that please?
Whiel using whisper (not whisperX) I obtained better sentences, smaller ones neither too long neither too small.
Thanks

[adamlogan]

My preferred way to reduce line length more naturally for subtitle output is to use:

--segment_resolution chunk

If people talk a lot, the lines can still be quite long though. You may still need to use:

--max_line_width MAX_LINE_WIDTH
(not possible with --no_align) the maximum number of characters in a line before breaking the line (default: None)
--max_line_count MAX_LINE_COUNT
(not possible with --no_align) the maximum number of lines in a segment (default: None)

I suggest trying with these values to start with:
--max_line_width 40
--max_line_count 2

You will probably still end up with orphan lines with just one or two words here and there, managing that, you'll probably need to use other advanced tools outside the scope of what WhisperX provides.

In the future, please try and consult the help files with CLI utilities, most have them by using flags -h or --help flags, a man page, or online documentation.

[dwids]

I had the same issue. I found this, which - for me - works a treat
https://github.com/peterk/srt_equalizer
My 'convert_audio_to_srt' Powershell script invokes WhisperX, then calls a small python script using srt_equalizer to tidy up the generated srt file. I used AI to write the python . Key line in it:
srt_equalizer.equalize_srt_file(input_srt, output_srt, 42)

[BestofthebestinAI]

Thanks interesting! How did you find this? (my post) Glad you did (did not try your solution yet but I like it already

[dwids]

I upgraded my WhisperX and it broke...so I was looking thru GitHub for solutions...and saw your question. As for my original issue, I went back and completely started again and got WhisperX working again. So my tip, particularly if you are using CUDA, is "if it ain't broke, don't upgrade!" 😊