Timings overlap problem with --max_line_width and --max_line_count options

[carolinaxxxxx]

Hi,

It seems that when using the --max_line_width and --max_line_count option there are incorrect timings in the output srt file:

Example:

1
00:00:15,912 --> 00:00:17,293
Good morning, loves! We're going to get 
breakfast! We're going to get breakfast at a new

2
00:00:19,935 --> 00:00:21,456
spot. It's called Spoon. Can't wait for you guys to 
see it. Enjoy! Please enjoy this Shed Eggs. Enjoy!

3
**00:01:02,962 --> 00:01:23,152**
hi guys oh hi sweethearts and welcome back to my 
channel welcome back to another video this is a

4
**00:01:02,962 --> 00:01:23,152**
highly anticipated video this is my birthday vlog 
thanks per welcome to my birthday vlog and it has

5
**00:01:02,962 --> 00:01:23,152**
been very eventful i'm not gonna lie am i too bright 
for you guys i probably am too bright let me

6
**00:01:02,962 --> 00:01:23,152**
probably like stand here welcome welcome welcome 
abroad welcome to thy sweet family welcome if

7
00:01:24,733 --> 00:01:46,482
you're new um if you're new to the channel i hope you 
like the video enough to give it a subscribe like

8
00:01:24,733 --> 00:01:46,482
down below and please comment guys i'm so happy i'm 
so sorry i need to calm down okay if you're new to the

9
00:01:47,842 --> 00:02:07,296
channel i hope you like the channel enough to give 
it a thumbs up and comment down below and leave a

10
00:01:47,842 --> 00:02:07,296
subscription to become part of this sweet family 
and become a sweetheart i hope that's more calm but

11
00:01:47,842 --> 00:02:07,296
yeah guys welcome to my birthday vlog we are about 
to okay wait i don't know what i'm saying about my

The times overlap **, so the output file is poorly constructed and, as a result, read.

Full command:

whisperx --batch_size 8 --model large-v3 --language en --device cuda --max_line_width 42 --max_line_count 2 --verbose False --output_format srt

It appears that using this option causes incorrect timestamps to be generated in the output files regardless of the model used. Without the option, timestamps are ok.

[carolinaxxxxx]

Can anyone confirm this issue? Thx.

[antonfp]

I have the same problem

[schemesmith]

same issue

[schemesmith]

oh actually was looking at the other issues and found a solution: set --highlight_words to True and just use some python script to get rid of the <u> </u>'s

[rubentorresbonet]

Same problem.