Speaker Anonymization
Speaker anonymization aims to remove identifying information from speech recordings while preserving linguistic content and other relevant features like emotion. Current research heavily utilizes deep learning models, particularly autoencoders and generative adversarial networks (GANs), often incorporating disentanglement techniques to separate speaker identity from other speech characteristics, with some work exploring signal processing methods and latent space transformations. This field is crucial for protecting user privacy in applications involving voice data, impacting areas like healthcare, voice assistants, and forensic speech analysis by enabling the use of speech data while mitigating privacy risks.
Papers
May 30, 2023
April 15, 2023
November 6, 2022
October 13, 2022
September 24, 2022
August 22, 2022
March 31, 2022
March 26, 2022
February 26, 2022