Qi-Yang Sun

ClASH (Controlled Lexical-Acoustic Separation Harness), a bilingual counterfactual diagnostic framework that evaluates each utterance under original, lexical-preserving, prosody-preserving, and approximately neutralised conditions, is introduced.

Imperceptible fingerprints prove to be far more reliable for speech deepfake attribution, even as perceptible ones fluctuate with emotional content and model changes.

Zero-shot LLMs can be significantly improved for sarcasm detection by integrating acoustic cues, achieving record-breaking accuracy without any model fine-tuning.

Adversarial attacks on Speech Emotion Recognition can now be both effective and interpretable, thanks to SIGMA's innovative saliency-guided approach.