Intelligibility and perceived quality of spectrally-impoverished speech sounds PACS: 43.71.Gv

Ten listeners were asked to identify and rate the quality of speech sounds processed in such a way that only the most energetic spectral components were retained. The influence of both the number of retained spectral components and of the width of the frequency bands surrounding these components was studied. The sounds consisted of vowel-consonant-vowel stimuli produced by two male and two female speakers. Retaining five channels containing the most energetic spectral components was generally sufficient for the sounds to be correctly identified, and twenty frequency channels with a 20-Hz width proved sufficient to produce a “correct” sound quality.