A binaural model for predicting speech intelligibility in rooms using noise and reverberation suppression processes

Abstract

Conventional metrics for predicting speech intelligibility are commonly described using the speech intelligibility index (SII) and speech transmission index (STI)--both of which are calculated using a monophonic signal. Under binaural conditions, these predictors often underestimate intelligibility due to the fact that beneficial binaural cues for unmasking detrimental effects are not accounted for. Beutelmann et al. [J. Acoust. Soc. Am. 127, 2479-2497 (2010)] proposed a binaural speech intelligibility model that determines the SII after suppressing spatially-separate noise using the equalization-cancellation (EC) process. This research expands on previous work by extending the model to analyze auditory scenes with more complex room acoustics, namely reverberation. In order to account for speech intelligibility degradation in the presence of room effects, the SII calculation is replaced by the STI. The current work also incorporates binaural release from reverberation by introducing an additional reverberation suppression mechanism into the model based on interaural coherence. The use of speech intelligibility metrics as an objective form of measurement may be used as a means to further understand binaural suppression processes within various room acoustic configurations through subjective listening tests.