ABSTRACT
This chapter presents a multiagent based sound stream segregation system for binaural input. Sound stream segregation is a technology used to extract individual sounds from a mixed sound, and it is considered to be primary processing for computationally understanding sounds (Computational Auditory Scene Analysis) in the real world. We already proposed residue-driven architecture as a computational model for a sound stream segregation. In this chapter, we discuss the design and implementation of two subsystems, called agencies, the BiHBSS agency and the BiGrouping agency, based on this architecture. The two agencies together segregate sound streams according to their harmonic structure and direction information. First, the BiHBSS agency segregates fragments of streams, and then, the BiGrouping agency groups the fragments to extract streams. Experimental results show that these agencies improve the quality of sound stream segregations.
