Modeling the mechanical aspects of the cochlea to determine acoustic phonetics

Using the sinusoidal model as developed by Quatieri and McAulay [IEEE Trans. Acoust. Speech Signal Process. ASSP-34, August (1986)] as an analysis tool that operates on both a time varying and frequency-dependent basis, a software design was developed to show how the inner ear resembles a series of bandpass filters. Each speech frame was divided into half-octave bands ranging from 125–4000 Hz. This was done by partitioning sinusoidal parameters based on frequency for the purpose of providing energy profiles. Speech analysis using the energy profiles of the bands was a ready method to determine formant locations for the purpose of identifying vowels, voiced/unvoiced regions, and an approximation of consonants. The software program runs on the TMS320C30 processing chip. Isolated phonemes were spoken into a microphone. For analysis purposes, the data obtained were compared to the corresponding band location of average formant frequencies as found in research conducted by Peterson and Barney [J. Acoust. Soc. Am. 24, 175–184 (1952)]. A more accurate analysis was given by comparing the software output of individual phonemes to the magnitude of the phoneme’s actual Fourier transform. The output of the software program compared almost perfectly.