Acoustic parameters for nasality based on a model of the auditory cortex

Abstract

Due to the longer length of the nasal tract as compared to the oral cavity, the asymmetry of the left and right nasal passages, and the coupling of the nasal tract to the sinuses, nasal sounds generally have more formants than oral sounds. However, all of the formants may not be prominent in the acoustic spectrum due to close-lying antiresonances. The net effect of the nasal peaks can be an increase in frequency modulation. In this work, we study the effectiveness of two new acoustic parameters (APs) that capture this increase in frequency modulation based on the ‘‘scale’’ dimension of a model of the auditory cortex [Chi et al., J. Acoust. Soc. Am. 118(2), 887–906 (2005)]. The new APs, mean center of gravity of scales and mean ratio of high scales to low scales, were used along with a previous set of four APs for the discrimination of nasal consonants and semivowels. Overall accuracy of 93.10% was obtained for this task. This corresponds to an 18.63% reduction in error rate over previously reported results. It remains to be seen if these APs are also useful in detecting nasalization in vowels. [Work supported by NSF Grant No. BCS0236707.]