Effect of voiced speech parameters on the intelligibility of PB words
Arun Agrawal, Wen C. Lin · The Journal of the Acoustical Society of America · 1975
The voiced/unvoiced regions of PB words are segmented on the basis of the energy in time segments. It is shown that the energy parameter works quite well and better than zero crossings rate (ZCR) for this segmentation. The real poles and formants of the voiced speech production model are obtained by solving a polynomial whose coefficients are determined by Prony’s or covariance formulation of linear prediction. The effect of these poles and formants on the intelligibility is investigated by using zero filters to suppress them adaptively. With this approach, one or more formants or real poles are removed from the natural speech without affecting anything else. The processed words were presented to listeners using an on−line speech intelligibility measurement system. The results indicated that the suppression of real poles and third and fourth formants did not degrade intelligibility, while the suppression of first and second formants lowered the average intelligibility scores by 7.6% and 27.6%, respectively. Subject Classification: 70.30.