[KongchangAI]
ConceptRecognizer Output Voting Error Reduction

ROVER

A classical ensemble algorithm for ASR systems. It constructs a confusion network through word-level alignment and selects the final word by voting, improving recognition robustness.

Timeline (last 90 days)

Nov 10

TID's seven-system ROVER ensemble achieved a WER of 36.25% on the unnormalized test set.

Unverified50%
Nov 10

TID's final submitted system combines seven models via the ROVER algorithm.

Unverified50%
Nov 10

ROVER first aligns the outputs of multiple systems at the word level to build a word confusion network, then at each aligned position selects the final word by voting according to occurrence frequency.

Unverified50%

All Facts (3)

Source Articles