Content constraints improve reversible speaker anonymization against unseen verifiers
On 38 unseen speakers, attacker equal error rate rose from 2.8% to 26.3%, with 11.8% word error rate.
Stories per day · 30 days
3 RSS
4 stories, newest first · one card per story
Stories tagged #speaker-verification
SPARC tests speaker linkage from speech transcripts alone
LLMs separate same-speaker pairs from text; 50-word probes weaken the strongest adversaries most.
Exploiting Acoustic and Content-Oriented Speaker Verification Attacks Against Multilingual Voice Anonymization
Acoustic ASV attackers generally beat content-oriented ones against multilingual voice anonymization.
Rethinking Automated Voice Similarity by Shifting from EER to Embedding Geometry
Voice-similarity alignment tracks embedding dimensionality, not verification EER; a bottleneck lifts it from 0.08 to 0.74.