[KongchangAI]
Conceptcounterfactual gain scorer

Counterfactual Gain Scorer

A scorer trained with soft labels constructed from differences between clean-audio and noisy decoding. It fuses acoustic and linguistic features to output token-level scores and aggregates them into segment-level scores for locating segments that require repetition.

Timeline (last 90 days)

Oct 20

\sysname trains a "counterfactual gain scorer" that fuses acoustic features and linguistic features, outputs token-level scores, and aggregates them into segment-level scores.

Unverified50%

All Facts (1)

Source Articles