eduzhai > Applied Sciences > Engineering >

Adversarial representation learning for private speech generation

  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: As more and more data is collected in various settings across organizations,companies, and countries, there has been an increase in the demand of userprivacy. Developing privacy preserving methods for data analytics is thus animportant area of research. In this work we present a model based on generativeadversarial networks (GANs) that learns to obfuscate specific sensitiveattributes in speech data. We train a model that learns to hide sensitiveinformation in the data, while preserving the meaning in the utterance. Themodel is trained in two steps: first to filter sensitive information in thespectrogram domain, and then to generate new and private informationindependent of the filtered one. The model is based on a U-Net CNN that takesmel-spectrograms as input. A MelGAN is used to invert the spectrograms back toraw audio waveforms. We show that it is possible to hide sensitive informationsuch as gender by generating new data, trained adversarially to maintainutility and realism.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...