subtitles too long

[livos]

Great project! Thank you. I just wonder why the subtitles are so long? They often show on 3 lines when generally subtitles show only on 2 lines max most part of the time. It takes too much space on the picture. Or maybe there is a way to set this ?

[absadiki]

You are welcome @livos :)

This depends on the model you are using. Some models provide a way to limit the length of each sentence, and some not.
For example you can choose ggerganov/whisper.cpp from the models, then from the configs you'll need to set token_timestamps to True and max_len to the maximum number of characters you want, this way you can control the maximum length of each subtitle.

Hope this helps ?

[livos]

Ok. I understand now. Thank you!

…

On Fri, Sep 15, 2023, 1:49 AM Abdeladim Sadiki ***@***.***> wrote: You are welcome @livos <https://github.com/livos> :) This depends on the model you are using. Some models provide a way to limit the length of each sentence, and some not. For example you can choose ggerganov/whisper.cpp from the models, then from the configs you'll need to set token_timestamps to True and max_len to the maximum number of characters you want, this way you can control the maximum length of each subtitle. Hope this helps ? — Reply to this email directly, view it on GitHub <#69 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AF4NBGMHF7ZTXFBUZ6OUSZ3X2NGTBANCNFSM6AAAAAA4YCHPWQ> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[raivisdejus]

I am using HuggingFace model to load a fine tuned whisper, everything in the tool works great, very useful and easy to use.

Generated subtitles show up word by word, I think it would be useful if we could combine the words into longer sections, with ability to specify some max length of the combined subtitle section. This could be a post processing tool on top of whatever whisper gives us.

[absadiki]

Good idea @raivisdejus!
so basically take the first segment and add the subsequent ones until you reach some max length ? right ?

[raivisdejus]

@abdeladim-s Best solution in real life is yet to be determined by testing. I did a proof of concept version of such "subtitle combiner" here https://gist.github.com/raivisdejus/07ca2e37d1fb87f81df12e424cf9175b (different setup, but similar approach)
This will combine words until it sees end of sentence punctuation ., ? or !. Max word limit or max subtitle segment length time wise are also options. I chose the end of sentence punctuation, for simplicity.

[absadiki]

Yes, a lot of approaches can be implemented.
Well, combining until punctuation is also good, if you can add this function as a post processing tool and submit a PR I'll be happy to merge it, otherwise whenever I find some free time I'll try to do it.