Public defence in Acoustics and Speech Technology, M.Sc.(Tech.) Pablo Pérez Zarazaga
M.Sc.(Tech.) Pablo Pérez Zarazaga will defend the thesis "Preserving Speech Privacy in Interactions with Ad Hoc Sensor Networks" on 28 October 2022 at 1 p.m. (EET) in Aalto University School of Electrical Engineering, Department of Signal Processing and Acoustics, in lecture hall F239a, Otakaari 3, Espoo, and online in Zoom.
Opponent: Prof. Rainer Martin, Ruhr-Universität Bochum, Germany
Custos: Prof. Tom Bäckström, Aalto University School of Electrical Engineering, Department of Signal Processing and Acoustics
Thesis available for public display 10 days prior to the defence at: https://aaltodoc.aalto.fi/doc_public/eonly/riiputus/
Doctoral theses in the School of Electrical Engineering: https://aaltodoc.aalto.fi/handle/123456789/53
Public defence announcement:
Speech is one of our most natural ways of communication, and voice user interfaces (VUIs) al-low us to communicate with our electronic devices in a natural way using our speech. For that reason, they have become popular applications, and their popularity is expected increase in the following years. We can find VUIs in a great variety of services, from communication applica-tions like Skype, to voice assistants like Alexa. In our everyday lives, we live surrounded by electronic devices that can provide a VUI and transfer data. Having many devices providing these services in the same environment would allow for collaboration between them. By sharing and jointly processing recorded signals between devices, it would be possible to improve the quality of the VUIs that each device offered individually. For example, using multi-channel signal processing to reduce the noise in a recorded conversation. However, our speech contains much of our personal information, and misusing a user’s voice information, like sharing their recorded speech with an unauthorized device, can cause a grave privacy violation. For that rea-son, in order to design collaborative systems between VUIs, it is necessary to define access management rules that protect the privacy of the users.
This thesis presents three main ideas: First, we assume that in our conversations with other people, we usually perceive a certain level of privacy, which affects the way we talk. Second, if our devices could recognize this perceived level of privacy, they could adapt their behavior to it. For example, only sharing our voice with devices that would naturally be able to hear our voice. Third, if the voice of an external user leaked into another conversation, their privacy would also be violated. Therefore, it is necessary to remove the voices of interfering speakers in our interactions with VUIs. Our work studies the effect of the environment on a speaker’s perception of privacy. We then apply this information to develop access management rules for VUIs that adapt to the user’s privacy requirements. For instance, if the user lowers the volume of their voice, access is restricted to devices outside the reach of the user’s voice. Finally, we present a method to remove speakers external to a conversation in a multi-device scenario.
Contact information of doctoral candidate: