Sound Scene Analysis with Psycho-acoustics Parameter

The sounds in our environments (environmental sound) contain many information, and we can often predict a surrounding situation from these sounds without visual information. Accurate sound scene analysis would be useful in various applications such as home security, surveillance and understanding of a human action and a natural environment. The feasibility of personal identification based on indoor footstep has been confirmed by using feature parameters extracted from footstep (Y. Shoji et al., 2004). We recorded typical environmental sounds generated by the car, the motorized bicycle(bike), the running footstep (running) and the walking footstep (walking), respectively. Recognition was examined using these experimental sounds. In this paper, we propose a k-means clustering using psycho-acoustic parameters for sound scene analysis. Results show a possibility that a surrounding situation can be predicted from environmental sounds