Predicting binaural gain in intelligibility and release from masking for speech.

Abstract

A simple, approximate procedure for predicting binaural gain in intelligibility and release from masking for detection of speech in broadband Gaussian noise is presented. The procedure assumes that the effect of binaural processing in the auditory system can be adequately represented by a frequency-dependent reduction in the level of the masking noise. The magnitude of this reduction is derived from data on release from masking for tones. Predictions of intelligibility are based on the Articulation Index as computed for the equivalent masking noise. It is predicted that the binaural gain in intelligibility for speech in white noise is greatest at low intelligibility levels, decreasing to about 3 dB at high intelligibility levels. The relative importance of low- and high-frequency interaural phase opposition in producing a binaural gain in intelligibility depends on the signal-to-noise (S/N) ratio. Low-frequency interaural phase opposition is of greater importance at low S/N ratios. Release from masking for detection represents a limiting case, the binaural gain being roughly 13 dB and dependent primarily on interaural phase information below about 500 Hz. Predictions based on this procedure are fairly consistent with data obtained in an earlier experiment. More data are required, however, for an adequate assessment of the technique. The implications for practical binaural systems are discussed.