Adversarial Representation Learning for Synthetic Replacement of Sensitive Speech Data

Examensarbete för masterexamen

Please use this identifier to cite or link to this item:
Download file(s):
File Description SizeFormat 
Master thesis Adam Östberg_ David Ericsson.pdf2.71 MBAdobe PDFThumbnail
Bibliographical item details
Type: Examensarbete för masterexamen
Title: Adversarial Representation Learning for Synthetic Replacement of Sensitive Speech Data
Authors: Östberg, Adam
Ericsson, David
Abstract: As more data is collected in various settings across organizations, companies, and countries, there has been an increase in the demand of user privacy. Developing privacy preserving methods for data analytics is thus an important area of research. In this work we present a model based on generative adversarial networks (GANs) that learns to obfuscate specific sensitive attributes in speech data. We train a model that learns to hide sensitive information in the data, while preserving the meaning in the utterance. The model is trained in two steps: first to filter sensitive information in the spectrogram domain, and then to generate new and private information independent of the filtered one. The model is based on a CNN that takes mel-spectrograms as input. A MelGAN is used to invert the spectrograms back to raw audio waveforms. We show that it is possible to hide sensitive information such as gender by generating new data, trained adversarially to maintain utility and realism.
Keywords: generative adversarial networks, adversarial representation learning, deep learning, privacy, speech generation
Issue Date: 2020
Publisher: Chalmers tekniska högskola / Institutionen för matematiska vetenskaper
Collection:Examensarbeten för masterexamen // Master Theses

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.