Model of binaural speech intelligibility in rooms.

Many models of speech intelligibility in rooms are based on monaural measures. However, the effect of binaural unmasking improves speech intelligibility substantially. The binaural speech intelligibility model (BSIM) uses multi-frequency-band equalization-cancellation (EC), which models human binaural noise reduction, and the Speech-Intelligibility-Index (SII), which calculates the resulting speech intelligibility. The model analyzes the signal-to-noise ratios at the left and the right ear (modeling better-ear-listening) and the interaural cross correlation of target speech and binaural interferer(s). The effect of the hearing threshold is modeled by assuming two uncorrelated threshold simulation noises for both ears. BSIM describes the (binaural) aspects of useful and detrimental room reflections, reverb, and background noise. Especially the interaction of delay time and direction of speech reflections with binaural unmasking in different acoustical situations was modeled successfully. BSIM can use either the binaural room impulse responses of speech and interferers together with their frequency spectra or binaural recordings of speech and noise. A short-term version of BSIM can be applied to modulated maskers and predicts the consequence of dip listening. Aspects of informational masking are not taken into account yet. To model different degrees of informational masking. the SII threshold has to be re-calibrated.