Perception of synthetic speech in quiet and in noise by adults with cochlear implants

Abstract

Recent research has demonstrated that the ability of adults with cochlear implants (CI) to perceive speech in noise is highly variable. There are many potential reasons why this may be so, including poor signal/noise segregation, and poor perception of phonetic features in noise. We examined the CI listeners’ perception of four synthetic speech continua in quiet and in noise. In two continua, dynamic spectral information was manipulated (/ra/-/la/ and /wa/-/ja/); in one, static spectral information was manipulated (/i/-/u/); and in one, only temporal information was manipulated (/sei/-/stei/). These were presented to 12 listeners with CIs and 13 normal-hearing listeners in quiet and concurrent with speech-shaped noise, at a +10 dB SNR. Noise affected identification (endpoints, boundaries, and slopes) of CI listeners more than NH listeners. Significant group-by-SNR interactions were found for endpoint identification of the /ra/-/la/, /wa/-/ja/, and /sei/-/stei/ continua. CI listeners had significantly shallower identification slopes for the /ra/-/la/ continuum; this was exaggerated in the +10 dB SNR condition. In addition, CI listeners showed more /u/ percepts than the NH listeners in the /i/-/u/ continuum at +10 dB SNR. Results are discussed with respect to the relative vulnerability of temporal and spectral features to misperception in noisy conditions.