Add optional TwelveLabs Pegasus transcription backend

[mohit-twelvelabs]

Hi! I'm Mohit, I work at TwelveLabs (@mohit-twelvelabs).

This adds an opt-in transcription backend to the notebook using TwelveLabs Pegasus, a video-understanding model, as an alternative to Faster Whisper.

What it adds

• A new markdown + code cell pair right after Run the model.
• Tick use_twelvelabs to transcribe with Pegasus; leave it unchecked and the notebook behaves exactly as before (Whisper output is kept).
• The Pegasus cell writes the same {start, end, text} fragments shape Whisper produces and sets language_detected, so the existing Merge and Translate cells run unchanged.

Why it helps this project

Pegasus watches the video server-side rather than only listening to the audio, so it tends to do better on on-screen text, speaker context and noisy audio — and it needs no local GPU, which is handy for users without an RTX card or who hit Colab GPU limits.

Opt-in / non-breaking

Defaults are untouched: use_twelvelabs = False. No existing cell, dependency or default behavior changes. The only new dependency (twelvelabs>=1.2.8) is installed lazily inside the opt-in cell.

How it was tested

• Verified the client.analyze(model_name=, video=VideoContext_Url(url=), prompt=, max_tokens=) signature and response.data field against the official SDK twelvelabs==1.2.8.
• Unit-checked the JSON-array extraction (handles prose/code-fence wrapping and drops empty segments).
• Live-smoke-tested the SDK + auth path against the TwelveLabs API (a Marengo text embedding returned a 512-dim vector). A full Pegasus run over a public video URL works server-side but can be slow, so the request wiring is what's asserted here.

You can grab a free API key at https://twelvelabs.io — there's a generous free tier.