Exploiting Acoustic and Content-Oriented Speaker Verification Attacks Against Multilingual Voice Anonymization
Acoustic ASV attackers generally beat content-oriented ones against multilingual voice anonymization.
The paper studies attacker automatic speaker verification against voice anonymization beyond English, comparing acoustic- and content-oriented attackers on multilingual anonymized speech. A multilingual voice-converted dataset, MultiVC, is built to improve cross-lingual generalization. Acoustic-oriented attackers generally perform better, but the gap narrows when linguistic information is well preserved. MultiVC further improves performance and partly reduces the cross-lingual gap; code, models, and the dataset are released.
- Attacker ASV on anonymized speech is tested in multilingual settings.
- Acoustic-oriented attackers outperform content-oriented attackers overall.
- The gap narrows when anonymized speech keeps linguistic utility.
- MultiVC improves attacker performance and reduces the cross-lingual gap.
From the original article
1 min read at arxiv.orgAttacker 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. Inspired by this, we investigate whether the same holds for attacker ASV on anonymized speech.
By Ridwan Arefeen · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.