Auditory model based direction estimation of concurrent speakers from binaural signals

Humans show a very robust ability to localize sounds in adverse conditions. Computational models of binaural sound localization and technical approaches of direction-of-arrival (DOA) estimation also show good performance, however, both their binaural feature extraction and the strategies for further analysis partly differ from what is currently known about the human auditory system. This study investigates auditory model based DOA estimation emphasizing known features and limitations of the auditory binaural processing such as (i) high temporal resolution, (ii) restricted frequency range to exploit temporal fine-structure, (iii) use of temporal envelope disparities, and (iv) a limited range to compensate for interaural time delay. DOA estimation performance was investigated for up to five concurrent speakers in free field and for up to three speakers in the presence of noise. The DOA errors in these conditions were always smaller than 5°. A condition with moving speakers was also tested and up to three moving speakers could be tracked simultaneously. Analysis of DOA performance as a function of the binaural temporal resolution showed that short time constants of about 5 ms employed by the auditory model were crucial for robustness against concurrent sources.