FEATURES AND NUMBER OF GAUSSIAN MIXTURES SELECTION FOR SINGING VOICE CLASSIFICATION IN COMMERCIAL MUSIC PRODUCTION

International audience The field of automatic classification of singing voices is still an open problem. In this context, we are interested with the classification and the assessment of singing voices. For this purpose, a sample database containing music files of Algerian singers is used. First, we make a separation between the voice and music parts in a song. Based on the vocal part, the some parameterswere extracted. This paper presents the parameters specifically designed for the analysis of a singing voice. A decision system was formed based on GMM (Gaussian mixture model); this system was used for classification of singing voice type and evaluation of singing voice quality. This paper presents our features selection task and the determination of the Gaussians number. Results show substantial improvements.