The effect of semantic cues on intelligibility: A comparison between spectrally sparse speech and natural speech in noise

Abstract

Listeners are known to make use of contextual cues when perceiving speech. Using contextual information is even more important when listening to distorted or difficult speech signals. For example, listeners benefit from contextual cues to compensate for the absence of fine acoustic-phonetic information when listening to natural speech in noise. In the present investigation, the effect of context was compared between four- and eight-channel spectrally degraded speech (Shannon et al., 1995) versus natural speech-in-noise (0 dB SNR). Spectrally degraded speech simulates the primary information transmitted by cochlear implant devices. The results demonstrated that eight-band signals received benefit from context to the same extent as natural speech in noise. In contrast, four-band signals provided significantly less benefit from context than natural speech in noise. The most parsimonious explanation of this pattern of results is that listeners need a threshold amount of acoustic information to make use of context equally, regardless of the type of distortion. Alternatively, the results could be explained by assuming that listeners employ different strategies depending on the overall acoustic characteristics of the speech signal. Future experiments will distinguish between these alternatives.