《AUDIO FEATURE EXTRACTION》.pdfVIP

HARMONIC-TEMPORAL-STRUCTURED CLUSTERING VIA DETERMINISTIC ANNEALING EM ALGORITHM FOR AUDIO FEATURE EXTRACTION Hirokazu Kameoka, Takuya Nishimoto, Shigeki Sagayama Graduate School of Information Science and Technology, The University of Tokyo 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8656, Japan e-mail: kameoka,nishi,sagayama@hil.t.u-tokyo.ac.jp ABSTRACT of the numerous conventional methods for multi-pitch analysis has been considered to be far from a practi- This paper proposes “harmonic-temporal structured clus- cal use. However, the recent pioneering ideas, e.g., tering (HTC) method”, that allows simultaneous estima- graphical model based (Kashino et al., 1995), filter- tion of pitch, intensity, onset, duration, etc., of each under- bank based (Klapuri et al., 2000), Kalman filtering based lying source in multi-stream audio signal, which we ex- (Nishi et al., 1996; Abe and Ando, 2000), multi-agent pect to be an effective feature extraction for MIR systems. based (Nakatani, 2002) and parametric signal and spec- STC decomposes the energy patterns diffused in time- trum modelings based approaches (Feder and Weinstein, frequency space, i.e., a time series of power spectrum, into