Binaurally integrated cross-correlation/auto-correlation mechanism (BICAM)

Abstract

A new precedence effect model is described that can use a binaural signal to robustly localize a sound source in the presence of multiple reflections. The model also extracts the delays (compared to the direct sound source) and lateral positions of each of the distinct reflections. A second-layer cross-correlation algorithm is introduced on top of a first layer autocorrelation/cross-correlation mechanism to determine the interaural time difference (ITD) of the direct sound source component. The ITD is then used to time align two auto-correlation functions obtained from the left and right ear signals to gather information about the reflections and form a binaural activity pattern. The model is able to simulate psychoacoustic lateralization results for a simulated direct sound source and a single reflection also for cases where the reflection exceeds the intensity of the direct sound (Haas Effect). Using head-related transfer functions to spatialize the sound sources, the model can accurately localize a speech signal in the presence of two or more early side reflections and late reverberation. The model can handle reverberation times of 2 s and above. [This material is based upon work supported by the National Science Foundation under Grant No. 1320059.]