Timbre classification method based on the Common Spatial Pattern filter

Abstract

The timbre of audio signals has not been clearly defined mathematically. It has been speculated that the time-frequency structure of audio signals is contributed to the timbre property. In this paper, we construct covariance matrix from the multi-band filter bank output signals of an audio signals and apply Common Spatial Pattern (CSP) filter to characterize timbre of audio signals. Simulation results confirms that the covariance matrix from the multi-band audio signals and CSP filter can be used as a potential feature of timbre classification.