An Algorithm Of Singer Identification System Based On GMM And Auditory Features

With the development of Internet technology, multimedia information resource is increasing rapidly. In the face of abundant music sources in the multimedia, people find it quite difficult to query the target music they need. Thus, how to realize the computer analysis and computer perception of the user’s music demand has become a future development objective of the human-computer interaction capability. Content-based music information retrieval is mainly applied on automatic music classification and recognition, and music genre classification and singer identification are relatively common among the multiple music classifications. Singer identification in the music automatic classification is an important branch of the music information retrieval. Singer identification uses the machine recognition method to simulate the experience of human ear to identify the singer’s voice.The paper focused on the application of Gaussian Mixture Model algorithm and the singer identification based on the auditory features of human ear, which belongs to the content-based music identification area. By using the human auditory features to identify a singer, the recognition accuracy has been improved and the content of music information retrieval has been enriched.This paper firstly discussed the basic features of music and voice, and mainly proposed that the auditory features could be employed as the recognition feature. Secondly, the paper studied the content of the singing voice detection in a song, and at the same time, the sparse representation algorithm was used as analysis. Thirdly, we emphasized the priority of the Gaussian mixture model algorithm in dealing with the speech signals recognition again and also proved its feasibility in singer identification field. Finally, experiment was done to verify the new algorithm of voice recognition in a song, and as a result the feasibility of the new algorithm was improved as well. On the other hand, the paper tested different results under different features attracted, and our effect was confirmed.