Any Whisper models specifically optimized or suitable for JAV?

C

Carlxlx

New Member
Dec 7, 2024
1
0
1
32
I got some JAV to watch, but havn't found any subtitles of them. I tried faster-whisper-large-v3-turbo to transcript them, but honestly the results aren't so good, they're just better than having nothing at all in my opinion.

I tried to google it, such as WhisperJAV, but they're just slightly better than original faster-whisper-large-v3-turbo, I still couldn't understand most of their words, all I can do is using my imagination.

So, are there any Whisper models specifically optimized or suitable for JAV, and better than WhisperJAV?

Thanks a lot
 
I've only tried on a couple of videos but I found that whisper large v2 did a lot better than v3 in terms of accuracy and not hallucinating random stuff as badly, although it still hallucinated terribly during quiet or musical portions of the test videos.
 
  • Like
Reactions: Carlxlx
Trying a huge variety of versions, the one that has given me the best results is r192.3 with the medium model (Identifies in an excellent way, the sentences) and with VAD Silero VAD 4.0 (the 5.0 fails to detect the voice)..

But if I want to increase quality especially in video with lots of music or background noise, you can first apply filters to separate the voice from the background.
github.com

GitHub - ZFTurbo/MVSEP-MDX23-music-separation-model: Model for MDX23 music separation contest

Model for MDX23 music separation contest. Contribute to ZFTurbo/MVSEP-MDX23-music-separation-model development by creating an account on GitHub.
github.com github.com

And with Gemini, chatpt, Llama, I do the translation I can get 80-95%.
 
You must log in or register to reply here.
Top Bottom