The research of audio clustering with Gaussian mixture based on EM algorithm
Yunhui Wang, Xiaoqing Yu, Wengen Wang, Liang Liu · 2011
In this paper, we present an approach to audio clustering, based on EM Algorithm with Gaussian Mixture. The proposed algorithm is simple and practical; it has an advantage in mass data processing. By improving it, the algorithm can be applied in audio MFCC feature clustering. For further exploration and research, firstly, we make a division of the library into speech and music by Zero-crossing Rate. And then, it is key point to further classify the library of music, such as pop music, rock music, and classical music and so on. In this process, we adopt Gaussian Mixture based on EM Algorithm, using 12-dimensional MFCC (Mel Frequency Cesptral Coefficient) as a feature vector set. The experimental results show that the proposed algorithm can demonstrate that the algorithm increases rate of audio classification compared with the unsupervised study and has good clustering ability.