Audio to text WhisperX

[madey83]

When i select polish language
it is not change to Polish, and transcription results are empty file.

[shoval6]

Same as you.
Any solution?

[darnn]

I can confirm many languages are missing in the language menu for WhisperX:

And here is my own log file:

whisper_log.txt

However, I do think removing it from the 4.X branch was the right idea. On Windows it's too resource-intensive (or was when I did manage to get it to work) and not really better than faster-whisper-xxl in any way.

If anyone is really dead set on using WhisperX specifically, I've been able to do it on Kaggle in the past. It's free to register and easy enough to set up.
Once you make a new notebook and upload your file, the commands you need are basically
!pip install git+https://github.com/m-bain/whisperx.git --root-user-action=ignore
and then
!whisperx --language en --model large-v2 --compute_type float32 --batch_size 3 --output_format srt --output_dir [outputdir] "/kaggle/input/datasets/[yourusername]/[folder]/[filename].wav"

With [folder] being what you named the folder when uploading your file, and [filename] being, naturally, the filename. I happened to use a wav file, but it takes any file that Whisper accepts as input.