Application of a model of the analog ear to speech signal analysis

Abstract

It is now a well-accepted fact that the transmission response of the human ear from the pressure incident at the pinna to displacement of the basilar membrane provides a means of resolving an incident acoustic signal into its spectrum energy distribution. However, due to the inherent nature of this transmission response, the resolution of the spectrum energy distribution that is produced is very poor, and is not at all in concert with the well-known capacity of human hearing for the perception of frequency. This paper reports on a system that is currently being experimentally pursued for enhancement of the resolution of the spectral energy transfer response of the basilar membrane to incident acoustical signals. The system consists of a 40-section electrical network analog of the basilar membrane displacement. The signal energy intercepted by each section is detected and then processed by a high-order differencing network. Differencing networks up to the sixth order are being investigated. Experimental evaluation shows that a second-order differencing network significantly enhances the peak structure of the pattern generated by the basilar membrane analog, and defines with reasonable resolution the formants of the input utterances. Higher order networks appear unnecessary and even detrimental.