repeated words, non-English language

[violetgoing]

Help me please, what should I turn on so that the words do not repeat? I tried to turn on VAD, language Russian 52 minutes

[violetgoing]

Pls help, idk what to do, If i do vad too high and the quality is lost, if too low and the words are repeated

[jhj0517]

Is there any background music on your audio?

[violetgoing]

No, there are louder and quieter voices.

[violetgoing]

There's also extraneous noises like wind.

[violetgoing]

By the way, why can't I select a model from my computer? Like @Const-me

[jhj0517]

extraneous noises

This is a whisper hallucination that happens when there's background music or noise in the audio.
I'm closing this because this is the same as #152, please continue there!

By the way, why can't I select a model from my computer?

I don't understand what you mean, but if you want to use custom model (like faster-whisper-large-v2-japanese-5k-steps) then you can put any faster-whisper model into WebUI-Path\models\Whisper\faster-whisper directory and use it.

[violetgoing]

can i install this? it large-v3 but in WebUI-Path\models\Whisper\faster-whisper i have only large-v1

[violetgoing]

@jhj0517

[jhj0517]

Yes, if you manually download the model and place it into the WebUI-Path\models\Whisper\faster-whisper,
It should work fine. Let me know if you encounter any errors

[violetgoing]

step 1

step 2

[violetgoing]

@jhj0517

[jhj0517]

This is related to #253 and I think it's a cuDNN & CUDA & torch version Incompatibility problem.

You have some options to resolve this:

1. If you're using CUDA versions other than 12.4, install 12.4.

2. Remove the venv directory and run install.bat again.

3. If you still face the error, follow Perfview's solution

Or alternatively, if you're finding these cumbersome, use Docker for the application.
You can see Running with Docker guide.
( You will need to install and run docker-desktop first.)