Based at the Phonetics Laboratory, Daniel D. Lee is a PhD Candidate in Computation, Cognition and Language. His doctoral research concerns the generation and detection of spoofed speech (also known as audio deepfake or voice cloning), which comprises both synthetic-based and imitation-based spoofing attacks. Undergirded by a 'four quadrant' approach (see figure below), the overarching objective is to investigate the acoustic, perceptual, and neurocognitive properties of spoofed speech.
While voice spoofing/anti-spoofing has predominantly been examined from the speech technology and engineering perspective, understanding the phonetic and phonological correlates of voice spoofing within each quadrant—as well as the overlaps and contrasts between quadrants—can offer deeper insights into the mechanisms involved in voice spoofing. Through the bridging of speech science and speech technology, both theoretical and applied findings promise to address real-world problems relating to voice spoofing.
Prior to coming to Cambridge, Daniel received his BA (Hons) and MA (by research) from Nanyang Technological University in Singapore, with a research focus on speech anatomy and physiology. His MA thesis examined voice quality (Laver, 1980) using real-time magnetic resonance imaging (rtMRI), revealing the theoretical relations between supralaryngeal settings and their impact on speech articulation. Among other findings, Daniel's voice quality experiments showed that vocal tract configurations (e.g. protruded mandible) influence segmental production in complex and asymmetric ways, guided by Laverian principles of compatibility and susceptibility. In Daniel's current work on spoofed speech, there are crucial implications for forensic speech science and cybersecurity. Similarly, his background in experimental voice quality research affords a unique lens through which one can view and evaluate what makes for a human-sounding 'voice'.
