Yi-Wen Liu




Updated in Apr 2021

My primary area is audio signal processing and the science of hearing. I am leading the Acoustics and Hearing Group (AHG) within the Graduate Institute of Electrical Engineering. Our long-term research topics include:

Modeling of inner ear physiology: from basic science to clinical applications

Mammalian hearing (including humans) is sensitive to signals as weak as 10e-17 Watt.  For many years, scientists have studied its mechanisms, and found that "outer hair cells", a special kind of sensory cells located inside the cochlea, are responsible for amplification of acoustic signals.   These cells can convert mechanical energy and electrical energy in both directions, thus forming a positive feedback loop that acts as an amplifier. However, it is still debatable how this electro-mechanical feedback happens exactly.   My long-term research interests, though kind of halted currently, is to construct computer-based dynamic-system models from protein, cellular, to system levels, thus simulating nonlinear responses of both normal and pathological ears. This research in particular leads to better understanding of what is called otoacoustic emissions (OAE), i.e. sounds coming from the ear, and OAE can be used as a non-invasive, quantitative tool for the diagnosis and prognosis of various sorts of auditory pathology. Another (more fruitful) research thread is to look at OAEs from purely signal-processing perpectives, and our current efforts include to estimate click-evoked OAE rigorously under noisy conditions, and to separate distortion-product OAE (DPOAE) from loudspeaker distortion. This research is currently collaborated with Dr. Hau-Tieng Wu (Math department, Duke University), Dr. PC Wang (ENT, Cathay General Hospital, Taipei), and Dr. Vincent Wang (ENT, China Medical University Hospital, Hsinchu). We seek industrial collaborations as well.

Speech signal processing

The AHG also has diverse interestss in audio and speech signal processing from the beginning, but our recent focus is the synthesis of voice signals. As of 2021, our group members keep innovating in two topics: Mandarin text-to-speech conversion, and Mandarin singing voice synthesis. Both topics of course leverage on recent development in deep learning. The project is funded by multiple government grants since 2018, and we continue to seek top talents and enthusiasts to join the lab in the next few years (say 2022-2025). Our main collaborator for the government-funded projects is Prof. Shan-Hung Wu of the CS department. Over the past few years, lab members have also been co-advised by colleagues in Academia Sinica. We currently are working with an industrial partner as well.

Music in the AI era

Prof. Yi-Wen Liu is an amateur pianist himself. We currently have one industrial project on automatic jamming with the guitar being the lead, and a project on analyzing and simulating synchronized behavior within the realm of classical music performance. We welcome students with musical passion x coding abilities to join the club; let us polish and explore any musically relevant ideas with technology. Together, we can become the CCRMA of NTHU.