As the use of speech as a modality of interacting with devices becomes more popular, a natural question arises -- can these systems be trusted? In addition to the many usual requirements of trustworthiness such as accuracy, reliability, fairness, security, etc., speech systems face unique challenges relating to privacy. Speech carries a lot more information than the mere content of what was spoken -- it also carries information about the speakers themselves, including their gender, nationality and native language, emotional state, health, and a variety of other biometric and demographic information. In the process of using the speech service, users expose themselves to unintended exploitation of this information. The privacy risks associated with the use of speech based interaces is now increasingly recognized by governments and corporate institutions alike.
In this talk we will discuss the privacy-related challenges of the use of speech as a biometric signal, and the need for protection. We will discuss the legal landscape of the problem, and introduce the various solutions that have been proposed, including cryptographic, multi-party-computation based, and hashing-based solutions, and their limitations.
Finally, time permitting we will briefly also discuss more recent methods based on adversarial processing of speech, and how these solutions may in fact serendipitously address some of the other aspects of trustworthiness of speech-processing.
Dr. Bhiksha Raj is a tenured (full) professor of Computer Science at Carnegie Mellon University. Dr. Raj completed his Ph.D in Electrical engineering and Computer Science from Carnegie Mellon University, USA, in 2000. He was at Compaq (Cambridge) Research Lab until 2001. From 2001 to 2008 he led Speech Research at Mitsubishi Electric Research Labs. Since 2008 he has been a full-time faculty at Carnegie Mellon. Over his career Dr. Raj has made pioneering contributions to three broad areas of research: Speech and Audio Processing, Privacy and Security in Multimedia Processing, and lately, Deep Learning and AI. He holds over 30 patents in these areas, is co-editor of three technical books and has published over over 360 research papers in peer-reviewed journals and conferences. His current research spans topics of high contemporary importance, such as exploiting data and structure redundancy for deep learning and AI systems, preserving user privacy in speech and audio processing systems, learning and evaluating classifiers under real-world labeling assumptions, and robustness of AI systems to adversarial attacks. He is a fellow of the IEEE.