Automatic Analysis And Research On Music Genre And Singer Voice

With the rapid development of computer and communication technology, the multimedia data on the Internet grows quickly. Music is a kind of the multimedia data, in order to organize these data quickly and effectively, we must investigate good method to classify the genres of music and the sound of the singers automaticly.This article mainly investigates the method of musical genre classification and singers'sound style classification. The main work is as follows:At first, we investigate the method of feature extraction in musical genre classification and singers'sound style classification. We propose feature extraction's means of CQT and beat histogram, CQT is a feature which is used for musical analysis specially, and beat histogram is a feature that captures the characteristic of musical rhythm. In addition, we also use the features that are commonly in signal processing, such as MFCC and its first-order and second-order difference, LPCC and its first-order and second-order difference, short-term energy and formants.Then we propose musical genre vector and radar chart which are used to provide a visual analysis. According to likelihood ratios of Gaussian mixture model, we establish musical genre vector and then plot the radar chart, which is used to investigate the misclassification degree between genres and provide a visual analysis.At last, we construct a system of musical genre and singers'sound style, and investigate them deeply. In aspect of musical genre, we study the music that belongs to one genre and dual-genre at the same time. By comparing the classification accuracy of short-term tembral features, we choose the best short-term features, then through feature fusion of short-term features and long-term features, a good performance is achieved over our database. We also discuss the influence on the classification accuracy when the number of Gaussian component changes and find the most appropriate Gaussian component. Meanwhile we investigate the accuracy when CQT is used for musical genre classification. In aspect of singers'sound style, we study the classification accuracy when MFCC and its first-order, second-order difference and formants are used for features.