Secrets Your Voice Reveals
Voice Recognition Attacks; IoT privacy
Millions of people have embraced voice services as an effortless way to communicate with smart home devices, smart vehicles, and other IoT devices due to their ease of use. However, voice analysis can reveal sensitive information about a person, including their gender, emotional state, and even health conditions. My research explores attribute obfuscation beyond speaker identity, targeting both audio and text domains. Specifically, I focus on Alzheimer’s disease, which can be identified by its speech and linguistic characteristics. I aim to convert attributes from dementia speech to healthy speech with the goal of rendering the disability undetectable while preserving privacy. This talk will explore the privacy risks associated with voice services and how we can combat them with privacy-preserving speech processing.