Shinnosuke Takamichi

Unigram entropy emerges as a key predictor of codec performance, revealing stark differences in how various architectures handle noise conditions.

Lowering bitrate settings in speech synthesis can still yield high-quality, intelligible output, challenging traditional assumptions about GSLM configurations.

ELSA outperforms existing metrics by aligning audio evaluations with human judgments at the level of individual acoustic events, transforming TTA assessment.