search

jianfch / stable-ts

Transcription, forced alignment, and audio indexing with OpenAI's Whisper
MIT License
1.61k stars 178 forks source link

request for outputs to be suitable for youtube shorts #95

Closed mikebilly closed 1 year ago

mikebilly commented 1 year ago

thanks for your project, I really like that it can caption down to word-level. I'm trying to make youtube shorts, using caption at one-word or 2-3-word at a time, and I don't want the word to disappear right after it has been spoken, but instead stay for a couple of miliseconds before disappearing or even stay on the screen until the new words appear. In short, I want to make captions suitable for youtube shorts

MilkAndBanana01 commented 1 year ago

I have been able to replicate a similar effect by using SubtitleEdit.

MilkAndBanana01 commented 1 year ago

@MilkAndBanana01 any bash commands to do so to make subtitle for shorts?

Currently, I don't have a console-only version of this workflow. For the most part, here's the workflow I have so far.

stable-ts input.mp4 --o input.srt --segment_level false

This pretty much spits out an srt file that has each word seperated with its own timestamp. Then you open SubtitleEdit and you can fill in the gaps between the words with Tools > Bridge Gaps Between Subtitles. image

You can also do Merge short lines... to get the 2-3 word effect that some Youtube Shorts does. image

I'm planning to see if I can automate this process so that when it runs stable-ts, it will automatically do certain parameters for the type of subtitle I want. I also have a potential in-built solution for this that will process the srt/txt file seperately to get certain subtitle effects.

Hope that helps!

jianfch commented 1 year ago

The latest version should let you cap each segment to a character/word count with split_by_length().