Application of sub-band spectral centroid features to recognizing underwater targets

Timbre is third attribute of subjective experience for a sound stimulus beside loudness and tone which distinguishes two sounds of equal loudness and tone. Timbre is hardly given by a value but described by physical parameters in a multidimensional space. The spectral centroid characters and their application to recognizing targets were studied. The spectral centroid, as one of physical parameters in timbre space, reflects sound frequency mean based on spectral energy distribution and is determined by several formant magnitudes and corresponding frequencies, that is found by studying spectral centroids of different music instrument sounds. The spectral centroid can reflect the area of sample energy concentration and recognize targets, but the recognized probability of one kind of underwater targets is lower in recognizing their steady-state sound singnals. A method of analyzing frequency bands dynamically based on spectral centroid was proposed by multidimensional sub-band spectral centroid to improve the target recognizing probabilities effectively.