Voice Anonymization by Multimodal Adaptive Noise

This is the course project report for Speech Language Processing in 2022 Winter.


Speaker anonymization has gained increasing popularity recently. This project aims at evaluating three methods of voice anonymization based on adding noises. This project quantitatively evaluates the effects of these methods. In addition, the equal error rate (in speaker verification) and the word error rate (in speech recognition) are computed to evaluate the efficacy of the anonymization methods. All analysis codes are open-sourced at https://github.com/ZiningZhu/CSC2518.