The research of digital media instrument recognition method based on the distribution overlapping degree of the Gaussian mixture model

Huayun Long

COMPUTER MODELLING & NEW TECHNOLOGIES 2013 17(5A) 68-72

Hunan University of technology Music College of HUT Hunan Zhuzhou 421007, China

This article mainly studies the musical instrument recognition in digital media based on audio. First of all, the features ar e studied in this paper. so in this article we choose Gaussian mixture model as models of the instruments, we use K-means algorithm to initialize Gaussian mixture model, and use EM algorithm to train Gaussian mixture mode l. based on this, through the study we find that the convergence of EM algorithm relates to the overlap of Gaussian mixture model, therefore, we use the overlap of Gaussian mixture model to determine the convergence of EM algorithm, and carry out in the traditional Chinese musical instrument recognition system, through the experiments we find that when we use the overlap of Gaussian mixture model to determine the convergence of EM algorithm, the recognition rate can be obviously improved.