Unified Speech Data Processing Framework for Speaker Privacy Protection and Fake Audio Detection

Period: October 2023 - March 2027
Project title: Japan Science and Technology Agency Strategic Basic Research Program (PRESTO)
Project number: JPMJPR23P9
URL: https://projectdb.jst.go.jp/grant/JST-PROJECT-23828637/

In this infodemic era, a voice data processing platform capable of detecting fake voices generated by AI and protecting personal information such as voiceprints is essential. However, existing fake voice detection and voiceprint protection anonymization technologies are designed independently, so they cannot be jointly optimized, and the levels of detection and protection cannot be theoretically demonstrated. This research integrates fake voice detection and voice anonymization in a deep learning framework based on the likelihood ratio, and theoretically improves the optimization criteria of the detection and anonymization models.