FIG. 1.

(Color online) Illustration of the effect of 64-channel vocoding versus the lower-resolution procedures of Ding et al. (2014) on envelopes within individual cochlear bands. (A) A histogram of the group-delay-adjusted squared normalized-correlation (i.e., variance explained) between the envelope in intact SiB and 64-channel vocoded SiB, which is used in this present study (i), and the eight-channel (ii) and four-channel (iii) vocoding of Ding et al. (2014) vocoded SiB. The histograms are across different speech sentences and 128 different cochlear bands equally spaced on an ERB-number scale (Glasberg and Moore, 1990) from 80 to 6000 Hz. The 64-channel vocoding clearly better preserves the within-band envelopes than either the eight- or four-channel procedures of Ding et al. (2014) in that the 64-channel procedure captures an additional variance of more than 20%. This disruption of within-band envelopes using their technique was observed despite replicating their result of 0.99 correlation for the band-summed envelope (i.e., the basis for their conclusion that their vocoding preserved speech envelopes). (B) An example envelope derived from SiB for the 1.5-kHz speech band for intact SiB, our 64-channel vocoding, and the better-resolution, eight-channel vocoding from Ding et al. (2014) to visualize how our procedure yields band-specific envelopes that more closely match those of intact SiB.