List your company

ResearchFormerly WizWand

UniSRM-BENCH

Benchmarks

TaskDatasetSOTA resultTrendResults

utterance-level pairwise preference judgement

UniSRM-BENCH T1

65.06Accuracy

12

multi-turn dialogue speech evaluation

UniSRM-BENCH T4

88.89Accuracy

10

scenario-aware style consistency preference (Chinese)

UniSRM-BENCH T3-Zh

91.3Accuracy

10

scenario-aware style consistency preference (English)

UniSRM-BENCH T3-En

85.61Accuracy

10

fine-grained speech quality scoring

UniSRM-BENCH T2

0.551PCC

9