fix(whisperx): use whisperx.diarize.DiarizationPipeline with token kwarg

[Anai-Guo]

What

Fixes the WhisperX diarization path, which currently fails with:

AttributeError: module 'whisperx' has no attribute 'DiarizationPipeline'

returning HTTP 200 with an empty text whenever a transcription is requested with diarize=true.

Why

The backend pins whisperx @ git+https://github.com/m-bain/whisperX.git (latest). Since WhisperX 3.6.0 the diarization API changed:

• DiarizationPipeline is no longer re-exported at the top level (whisperx/__init__.py only lazily exposes load_model, load_audio, align, load_align_model, assign_word_speakers, ...). It now lives in whisperx.diarize.
• Its constructor kwarg was renamed use_auth_token -> token (DiarizationPipeline(model_name=None, token=None, device="cpu", cache_dir=None)).

Change

• Import DiarizationPipeline from whisperx.diarize.
• Pass the HF token via token= instead of use_auth_token=.

whisperx.assign_word_speakers(...) is still exported, so it is left unchanged. Two-line change, no behavior change beyond making diarization work again.

Fixes #10388

AI-assisted, human reviewed.

🤖 Generated with Claude Code

[mudler]

would be nice to have tests catching that, but it puts strain on the CI to test everything - thanks for looking into it