fix: handle pyannote legacy and pre-4.x diarization return types

[NotYuSheng]

Summary

• diarization_service.py assumed pyannote 4.x always returns a DiarizeOutput object (accessing .speaker_diarization)
• This breaks when legacy=True is set on the pipeline, or if someone runs with pyannote <4.0 (which returns an Annotation directly)
• Added a hasattr guard to handle both return types gracefully

Root cause

pyannote 4.x SpeakerDiarization.apply() returns:

• DiarizeOutput when legacy=False (default)
• Annotation directly when legacy=True (backwards compat mode)

Test plan

• Upload a file and confirm diarization completes with speaker labels
• Confirm behaviour is unchanged from current pyannote 4.0.4 with legacy=False

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request updates the diarization service to support both legacy and modern PyAnnote return types by conditionally accessing the speaker diarization attribute. The review feedback suggests improving the implementation's idiomaticity by using getattr and adding a safety check for None values to prevent potential runtime errors.

I am having trouble creating individual review comments. Click here to see my feedback.

backend/services/diarization_service.py (87-91)

The current implementation uses a verbose ternary operator to handle the different return types from PyAnnote. Using getattr with a default value is more idiomatic in Python for this pattern. Additionally, adding an explicit check for None ensures the service handles cases where the pipeline might fail to return a result without raising a confusing AttributeError when itertracks is called later.

            if diarization is None:
                raise ValueError("Diarization pipeline returned no result")
            annotation = getattr(diarization, "speaker_diarization", diarization)

[NotYuSheng]

/gemini review

[gemini-code-assist]

Code Review

This pull request updates a comment in the backend diarization service and enhances the speaker badge UI in the frontend with flexbox layout and text truncation. Feedback highlights a missing logic implementation in the backend to handle different pyannote return types, which could cause an AttributeError. In the frontend, it is recommended to use the Bootstrap 'text-truncate' utility class instead of multiple inline styles for better maintainability.