Filterbank-based blind signal separation with estimated sound direction

Based on a mathematical model of the human auditory pathway, a binaural auditory system has been developed for sound localization and speech enhancement. The developed system includes a filterbank, and the blind signal separation is conducted at each filtered signal. The estimated directions of arrival are collected at each subband, and are utilized to localize the multiple sources from convolved mixtures. Imposing an additional constraint based on the estimated sound directions shows faster convergence; also, better signal separation is expected for noisy multi-source speech.