Play clips, compare methods, and record human feedback against the validation set
2 APR 2026
I. Validation Set
This review set is selected by variance axes, not random clips: speaker topology, pause profile, language mix, lexical challenge, recording condition, and domain context. The purpose is to expose different failure modes in Cantonese transcription.
Status
Current page loads whatever outputs already exist and marks missing methods explicitly. Feedback is stored in your browser and can be exported as JSON.