Adversarial optimization decouples UTMOS predictions from perceived speech quality in multiple input spaces, exposing failure modes of this widely used SQA model.
Good Practices for Evaluation of Synthesized Speech,
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
UNVERDICTED 2representative citing papers
An experiment shows humans detect fully synthetic speech below chance but exhibit implicit discrimination via quality ratings in a localization task with trust cue manipulations.
citing papers explorer
-
Attacking UTMOS: Probing the Robustness of a Speech Quality Assessment Model
Adversarial optimization decouples UTMOS predictions from perceived speech quality in multiple input spaces, exposing failure modes of this widely used SQA model.