Subcritical width rectangular bands of vocoded speech reveal the nature of envelope processing

Abstract

Bandwidth requirements for temporal envelope processing were examined using sentences that were reduced to arrays of sixteen rectangular bands (4800 dB/octave rolloff) having center frequencies ranging from 250 Hz to 8000 Hz, spaced at 1/3-octave intervals, and having subcritical bandwidths ranging from 4% to 0.5%. Envelopes extracted from the speech bands, without smoothing, were used to modulate white noise or tones centered at the speech band center frequencies. When the bandwidth of the modulated signals matched that of the speech, tone-vocoded arrays were more intelligible than noise-vocoded arrays and less intelligible than parent speech-band arrays. However, doubling the bandwidths of tone-vocoded arrays, which passed the upper and lower modulation sidebands fully, increased intelligibility to that of the parent speech array. Moreover, when the widths of modulated noise-band arrays were expanded to either an ERBn or 1/3-octave, their intelligibility equaled that of the speech for parent bandwidths of 4% and 2% and greatly exceeded speech array intelligibilities for parent bandwidths of 1% (58% vs. 25%) and 0.5% (28% vs. 3%). These and other findings indicate that optimal temporal envelope processing of speech requires that envelope cues stimulate a majority of fibers comprising critical bands. [Work supported by NIH.]