1. CRYPTOGRAPHY
For subjects related to mathematical aspects and post-quantum cryptography : contact Prof. Christophe Petit <christophe.petit@ulb.be<> (https://christophe.petit.web.ulb.be/ for research interests and https://christophe.petit.web.ulb.be/supervision.html for previous master thesis topics).
For subjects related to symmetric cyryptography : contact Prof. Rachelle Heim <rachelle.heim@ulb.be>.
2. NETWORK SECURITY
For subjects related to network security, physical-layer security, IoT security, SDN security, … : contact Prof. Jean-Michel Dricot <jean-michel.dricot@ulb.be>.
3. SECURE PROTOCOLS
For subjects related to the design and analysis of secure protocols as well as secure protocols for privacy : contact Prof. Olivier Markowitch <olivier.markowitch@ulb.be>.
Subject proposition : Privacy in voice assistants
Voice assistants (e.g., Google Assistant, Alexa…) are becoming ubiquitous. Unfortunately, the voice itself carries a lot of sensitive information. For instance, analysis of the voice can reveal information on the speaker such as their identity, age, sexe, emotions, mental state (e.g., drunk), health information (e.g., Parkinson disease, CoVid-19 infection…), origin (e.g., accent)…
On the bright side, researchers have proposed different techniques to improve the privacy of the speakers by removing sensitive information. However, those techniques remain empirical since it is still an open question which part of the speech contains the utility (which corresponds, in our context, to the underlying sentence of the request). Therefore, sanitized speech is usually accompanied by a quality drop in its utility. Furthermore, voice sanitization techniques often focus only on some sensitive information.
The objective of this master thesis is to combine several sanitization methods for different sensitive information and study their impact on the speech utility to find a good way to combine them. Since there exist a lot of sanitization techniques and kinds of sensitive information, we only expect the student to pick a few sensitive information to sanitize and their related state of the art sanitization methods to combine together.
What is expected concretely:
- Pick some kinds of sensitive information to sanitize (e.g., disease and age);
- Verify the state of the art for voice sanitization related to them (including combining them, if it exists);
- For the master thesis: test different combinations and analyse their impact (e.g., using the VoicePrivacy Challenge framework for voice utility and some models to detect those sensitive information).
Supervision meetings will take place online between February and June.
N.B. State of the art voice sanitization techniques mostly rely on different machine learning models so knowledge in machine learning is helpful.
Supervisor: Prof. Markowitch
Co-supervisor: Arnaud Leponce
4. HARDWARE SECURITY
For subjects related to trusted platforms and to hardware security: contact Prof. Jan Tobias Müelhberg <jan.tobias.muehlberg@ulb.be>