Real-Time Interaural Time Delay Estimation Via Onset Detection

Reliable real-time estimation of the interaural time delay of a sound source is difficult in the presence of noise and reverberation. However, the psychoacoustical precedence effect suggests that accurate estimation is possible by concentrating on the first-arriving sound. This paper introduces a novel real-time estimation method inspired by the precedence effect. First, the arrival of first-arriving sound is detected by performing a hypothesis test based on local approximation of the binaural signal with exponentially decaying sinusoids, which effectively model the shape of a sound onset. After detection, the interaural time delay is directly retrieved from the phase shift of the approximating sinusoids. The local model approximation is done with efficient recursions by parameterization of the model with autonomous linear state-space models, making the algorithm implementable in real-time.