Multi-Agent Based Binaural Sound Stream Segregation

Tomohiro Nakatani, Hiroshi G. Okuno and Takeshi Kawabata
(NTT Basic Research Laboratories)

This paper presents a multi-agent based sound stream
segregation system for binaural input. Sound stream
segregation is a technology to extract individual sounds
from a mixed sound, and it is considered as a primary
processing for computationally understanding sounds
(Computational Auditory Scene Analysis) in the real-world. 
We have already proposed the Residue-Driven Architecture as
a computational model for a sound stream segregation.  In
this paper, we discuss the design and implementation of two
subsystems, called agencies, the Bi-HBSS (Binaural-Harmonic
Based Sound stream Segregation) agency and the Bi-Groupping
agency, based on this architecture. Two agencies together
segregate sound streams according to their harmonic
structure and spatial location in parallel.  First, the
Bi-HBSS agency segregates fragments of streams, and then,
the Bi-Grouping agency group the fragments to extract
streams.  Experimental results show that these agencies can
segregate two sound streams using harmonic structure and
spatial location.