Relative contribution of temporal cues and spectral profile to voice gender discrimination

Abstract

Despite limited access to spectral and temporal cues, cochlear implant (CI) users are somewhat able to discriminate voice gender. The present study explored the relative contributions of spectral and temporal cues to normal-hearing (NH) subjects’ voice gender discrimination, while listening to multi-channel simulations of CI processing. The output spectrum was either matched (relative to normal) or upwardly shifted to simulate the spectral shift associated with CIs; the envelope filter in each channel was varied to examine the contribution of temporal cues. Voice gender discrimination was tested with two talker sets, in which the mean fundamental frequency (F0) between male and female talkers was either widely or narrowly separated. Results showed that for both talker sets, 16 spectral channels were needed before subjects could use the spectral profile to identify voice gender; when the speech spectrum was shifted, 32 channels were needed. Given enough temporal cues, the spectral profile had a relatively small effect on discrimination when the F0 was widely separated between male and female talkers. When there was little F0 separation between male and female talkers, the spectral profile had a much stronger effect; however, 16 or more channels were needed before listeners could attend to the profile. [Work supported by NIDCD-RO1-DC004993.]