Relative contributions of passband and filter skirts to intelligibility of bandpass speech: Some effects of amplitude and context

Abstract

Recently, Warren and Bashford [J. Acoust. Soc. Am. 106, L47–L52 (1999)] reported over 90% intelligibility for everyday sentences reduced to a 1/3-octave band (center frequency 1500 Hz, slopes 100 dB/octave, peak levels 75 dB). Surprisingly, when the bandpass sentences were partitioned using 2000-order FIR filtering, the rectangular 1/3-octave passbands had only 24% intelligibility, while the filter skirts separated by a 1/3-octave notch had an intelligibility of 83%, despite their severe spectral tilts. The present study found that when monosyllabic word lists were substituted for the sentences, the whole-band intelligibility was 26%, the passband 4%, and the filter skirts 16%. Intelligibility also was measured for 1/3-octave sentences with peak levels ranging from 85 down to 35 dB: The whole band intelligibilities ranged from 90% to 68%, and the filter skirt pairs each had three to four times the passbands’ intelligibility. These findings demonstrate the ability to compensate for extreme distortions of the spectral profile of speech. Hence, it is necessary to consider the contribution of frequencies outside the nominal passband as defined by the 3 dB downpoints along the filter slopes when dealing with filtered speech. [Work supported by NIH.]