update whisperx to 3.8.1

[bgruening]

No description provided.

[bgruening]

@arash77 any idea why the test fails?

Or better how could the test pass in your first version :)

[arash77]

@bgruening I can't remember how did they pass the test before! but should we include the models in the docker file so that the CI can access it?

[bgruening]

No, they are too large I think. Our models are here: https://github.com/usegalaxy-eu/infrastructure-playbook/blob/master/files/galaxy/tpv/tools.yml#L523

[arash77]

Previously, the test for this tool might not have even run because we didn't pass WHISPERX_MODEL_DIR to it. How can we test the tool in CI if we don't have access to the models? Should we use the expect_failure approach?

[bgruening]

I don't know, that was my questino. The CI def. run ... but it turned green. Can it be that huggingface downloaded the model on the fly?

[arash77]

But it needs an hf_token that has accepted the terms and conditions to download the models. Since we are using --model_cache_only True, it shouldn't be downloading them anyway.

[arash77]

Also by upgrading WhisperX, it now requires another model: pyannote/speaker-diarization-community-1.

[bgruening]

Ok, then I have no clue how this ever worked on CI :)

Do you have time to test it locally, and then we merge? Or should we YOLO?

[arash77]

I'm trying to test it, but it requires a GPU! I'm also not sure if the GPU is enabled on GitHub Actions but I see we fallback to cpu here.

[arash77]

There are some models that need to be added based on the WhisperX documentation, then we can run it on galaxy to test if it works.
pyannote/speaker-diarization-3.1 needs to change to pyannote/speaker-diarization-community-1
The NLTK punkt_tab tokenizer for alignment is also required.