Speech intelligibility and masking release using temporal fine structure and recovered envelope cues for normal-hearing and hearing-impaired listeners

Abstract

Consonant-identification ability was examined in normal-hearing (NH) and hearing-impaired (HI) listeners in the presence of continuous and 10-Hz square-wave interrupted speech-shaped noise. The speech stimuli (16 consonants in a-C-a syllables) were processed to present envelope (ENV) cues, temporal fine-structure cues (TFS), or envelope cues recovered from TFS speech (RENV). ENV- and TFS-speech was generated by extracting the ENV or TFS component of the unprocessed speech in N adjacent bands (N = 40 for ENV-speech, N = 1 or 4 for TFS-speech; range = 80–8020 Hz). RENV-speech was generated by extracting the ENV component of both types of TFS-speech in 40 adjacent bands. NH listeners were tested at an SNR of −10 dB and individual HI listeners were tested with SNR in the range of −6 to + 5 dB. These values of SNR yielded consonant-identification scores of roughly 50%-correct for intact speech in continuous noise for each listener. HI listeners had poorer speech scores than NH listeners. For both groups, scores with TFS- and RENV-speech were very similar. Scores were higher in interrupted noise than in continuous noise (indicating substantial release from masking), except for unprocessed- and ENV-speech for HI listeners. Audibility, frequency selectivity, and forward masking were estimated for each listener and compared with speech identification.