MACC'97 Session: Probabilistic Model in Multiagent Systems
We address a sound recognition problem where multiple acoustic objects are simultaneously present. The specific feature of this problem is that noise and signals are not predetermined: there may be many signals simultaneously. Therefore we consider a multiagent system architecture for this problem. Our first example is ensemble music recognition. Firstly, we propose methods for arbitration between agents and for probabilistic information integration. These methods are implemented in an ensemble music recognition system. We then discuss possible extention of the system to a general acoustic object recognition system.
Bayesian network, music recognition, auditory scene analysis
PS
file(+gzip) (in Japanese)