Video Recordings

For many years the human auditory system has been an inspiration for developers of automatic speech recognition systems because of its ability to interpret speech accurately in a wide variety of difficult acoustical environments. This talk will discuss the application of physiologically-motivated and psychophysically-motivated approaches to signal processing that facilitates robust automatic speech recognition. The talk will begin by reviewing selected aspects of auditory processing that are believed to be especially relevant to speech perception, and that had been components of signal processing schemes that were proposed in the 1980s. We will review and discuss the motivation for, and the structure of, classical and contemporary computational models of auditory processing that have been applied to speech recognition, and we will evaluate and compare their impact on improving speech recognition accuracy. We will discuss some of the general observations and results that have been obtained during the renaissance of activity in auditory-based features over the past 15 years. Finally, we will identify certain attributes of auditory processing that we believe to be generally helpful, and share insights that we have gleaned about auditory processing from recent work at Carnegie Mellon.