speaker-anonymization

This paper investigates how multimodal speaker verification using audio, prosodic, and linguistic cues across multiple utterances can significantly degrade speaker anonymization, reducing equal error rate by over 15% with only five utterances.