[bug] Diarization error

[Chewbacca96]

What happened
Getting error on diarization step, after successful transcribation.

Command / how you invoked it

./scripts/transcribe.sh '/path/to/audio' --lang ru --speakers 6

Expected
Correct operation of the script

Environment

• Chip / RAM: AMD 7800x3d / 64 gb
• OS: Ubuntu 24.04 on WSL
• scriba commit: latest
• Agent (if any): -

Logs

WARN: diarization failed (TypeError: Pipeline.from_pretrained() got an unexpected keyword argument 'token'); continuing without speaker labels

if i change in scripts/transcribe_whisperx.py:266 "token" to "use_auth_token", then i'm getting

WARN: diarization failed (TypeError: SpeakerDiarization.__init__() got an unexpected keyword argument 'plda'); continuing without speaker labels

[AlexanderAbramovPav]

Thanks for the clear repro — root cause found.

whisply (which scriba's bootstrap installs) hard-pins pyannote.audio==3.4.0, but scriba's diarization loads the pyannote/speaker-diarization-community-1 pipeline with token=, which needs pyannote.audio ≥ 4.0. On 3.x the arg is use_auth_token (your first error), and community-1's PLDA clustering doesn't exist (the plda error after your patch). It worked on my machine only because my venv was manually upgraded to the 4.x baseline — the bootstrap never did that upgrade, so fresh installs like yours broke. My fault, sorry for the friction.

Immediate fix on your end:

uv pip install --python skills/scriba/.venv/bin/python -U "whisperx>=3.8.6" "pyannote.audio>=4.0"

Then re-run your command — transcription was always fine; this just restores speaker labels. (No need to edit transcribe_whisperx.py; revert your use_auth_token change.)

Permanent fix: #5 makes the bootstrap install the pyannote.audio>=4.0 / whisperx>=3.8.6 baseline and self-heal existing venvs, with a clear preflight message if an old pyannote is ever present. Re-running bash scripts/transcribe.sh --bootstrap will upgrade your venv automatically once it lands.

Will close when the PR merges. Thanks again!