Modelling context processing during sentence recognition in noise and reverberation for listeners with and without hearing loss

This study evaluated three models quantifying context processing in human speech recognition. We presented everyday sentences with high semantic predictability and closed word set (matrix) sentences with low semantic predictability to listeners with normal hearing and listeners with hearing loss. Sentences were presented in quiet, in steady-state noise, and in speech-like modulated noise. Reverberated and clean speech were used. We evaluated whether the models are suited for testing the following hypotheses: (1) "Listeners with hearing loss compensate reduced auditory input by relying more strongly on speech context." (2) "Speech context is used to compensate missing information differently in modulated noise than in steady state noise or in quiet, with a smaller difference in context effects for listeners with hearing loss." (3) "Context processing explains why matrix sentences are much more robust against reverberation than everyday sentences." Significance testing was performed using bootstrapping. Only one of the context models showed significant differences in context processing between normal-hearing and hearing-impaired listeners for the different sentence types and for different listening conditions. No differences in context processing were found between stationary and modulated noise. To explain the behavior in reverberation, it is important to distinguish between a priori context and a posteriori context.