Phonetic cues are weighted differently when spectral resolution is degraded

Two experiments were conducted to test the hypothesis that listeners would modify perceptual strategies for phonetic identification if the spectrum of a speech signal was degraded. The contrasts between tense and lax vowels and between word-final s/z were selected because of the primacy of spectral cues and redundancy of temporal cues used in normal listening situations. In the first experiment, synthesized speech signals with vowels varying in two spectral dimensions (initial formant structure and vowel-inherent spectral change) and in one temporal dimension (vowel duration) were presented with intact spectra or through an 8or 4-channel noise-excited vocoder, used to simulate a cochlear implant. In the second experiment, similar manipulations were made to vowels and final consonants in a natural utterance, with variables of F1 offset transition, vowel duration, fricative duration and duration of consonant voicing. Identification results showed that as the spectrum was degraded, listeners relied less upon the spectral cues, and compensated with greater reliance on the temporal cues. Preliminary results with cochlear implant users suggest that results from the simulations are generalizable to a target clinical population; those who experience spectral degradation may employ perceptual strategies that differ from those used by individuals with normal hearing.