Exploiting Acoustic and Content-Oriented Speaker Verification Attacks Against Multilingual Voice Anonymization

Key points

• Conventional ASV has shown that both acoustic and contextual information are important for multilingual speaker verification.1
• Inspired by this, we investigate whether the same holds for attacker ASV on anonymized speech.1
• We evaluate both acoustic- and content-oriented attackers on multilingual anonymized speech and construct a multilingual voice-converted dataset to improve cross-lingual generalization.1
• Our results show that attacker effectiveness depends on the linguistic utility of the anonymized speech.1

Sources (1)

• [1]Exploiting Acoustic and Content-Oriented Speaker Verification Attacks Against Multilingual Voice Anonymization
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 6, 10:31 AM

  Attacker ASV systems for voice anonymization have been studied primarily in English, leaving their behavior in multilingual settings largely unexplored.

  Conventional ASV has shown that both acoustic and contextual information are important for multilingual speaker verification.

Extractive summary: sentences quoted from the sources.