Rebuild the demo around per-turn confidence

Publish the datasets' own speakers as a Reference pipeline

Shrink the example cards to a title plus a one-line subheadline

Reorder examples: LGC first, couch interview first in the real-world row

Cut demo to three 30 s excerpts and add three real-world examples

Remove internal pipeline chain details from UI and data

Deploy Multi-Voice demo

Deploy Multi-Voice demo

Deploy Multi-Voice demo