Features Based on Auditory Physiology and Perception

It is well known that human speech processing capabilities far surpass the capabilitiesof current automatic speech recognition and related technologies, despite very intensiveresearch in automated speech technologies in recent decades. Indeed, since the early 1980’s,this observation has motivated the development of speech recognition feature extractionapproaches that are inspired by auditory processing and perception, but it is only relativelyrecently that these approaches have become effective in their application to computer speechprocessing. The goal of this chapter is to review some of the major ways in which featureextraction schemes based on auditory processing have facilitated greater speech recognitionaccuracy in recent years, as well as to provide some insight into the nature of current trendsand future directions in this area.We begin this chapter with a brief review of some of the major physiological and perceptualphenomena that have motivated feature extraction algorithms based on auditory processing.We continue with a review and discussion of three seminal ‘classical’ auditory models of the1980s that have had a major impact on the approaches taken by more recent contributors tothis field. Finally, we turn our attention to selected more recent topics of interest in auditoryfeature analysis, along with some of the feature extraction approaches that have been basedon them.We conclude with a discussion of the attributes of auditory models that appear to bemost effective in improving speech recognition accuracy in difficult acoustic environments.

R. M. Stern and N. Morgan

None