Semantic representations involvement during degraded speech perception
The perceptual clarity of speech is not entirely dependent on the acoustic quality of the sound. Other resources, such as linguistic representations, are involved during degraded speech perception. For example, presentation of the written version of a degraded sentence before hearing it will enable prior knowledge on the exact speech content, which will make the degraded sentence seems clearer. This phenomenon has been explained by top-down influence of phonological and lexical representations on acoustic processing. Another example is the influence of semantic representations on the intelligibility of degraded speech: degraded sentences are better reported if they are meaningful than meaningless. The question now is whether the semantic representations could further influence the perceptual clarity of degraded speech for both normal-hearing (NH) and hearing-impaired (HI) listeners. In the reported set of three experiments, grammatically correct Swedish spoken sentences were presented at different sound quality levels, from clear to unintelligible. The sound quality levels were manipulated by using noise vocoding (NV) method in which the number of bands reflects intelligibility: more bands for more intelligibility. HI listeners were provided with amplification according to the Cambridge formula. The sentences had either high (e.g. “His new clothes were from France.”) or low (e.g. “His red school was from the newspaper”) semantic coherence and were matched at the word level. The written version of each spoken word (matching text) or a string of consonants (non-matching text) was presented 200 ms beforehand in a rapid serial visual paradigm. The task of the listeners was to rate the clarity of each spoken sentence on a 7-point Likert scale. Results revealed significant interactions between coherence and text for both groups, showing a benefit of coherence with matching and non-matching text for NH listeners but only with matching text for HI listeners. Significant three-way interactions including sound quality level modified this finding to some extent. Indeed, NH listeners benefitted from semantic coherence with non-matching text at 6 and 12 band NV (but not 3 band) while HI listeners benefitted at 12 band (but not 3 and 6 band). Preliminary fMRI results obtained for NH listeners indicated that processing of semantic coherence with non-matching text is supported by right middle temporal gyrus. The overall pattern of results suggest that NH listeners successfully utilize semantic representations in spoken sentences that are moderately degraded and when no prior knowledge is available. What prevents HI listeners to do the same?