Fine-tuning an SSL content encoder on Japanese, especially when the encoder is pre-trained multilingually, makes anonymized Japanese and Mandarin speech much more intelligible while keeping speaker privacy at usable levels.
Experiment Setup To assess the anonymizer’s ability to protect speaker identity, we adopted the privacy metric and evaluation methods from VPC
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
eess.AS 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Mitigating Language Mismatch in SSL-Based Speaker Anonymization
Fine-tuning an SSL content encoder on Japanese, especially when the encoder is pre-trained multilingually, makes anonymized Japanese and Mandarin speech much more intelligible while keeping speaker privacy at usable levels.