Research On An Algorithm Of Singer Recognition Based On GMM And Cepstrum Transformation

Today, multimedia resources on the Internet continue to increase, accompanied bythe rapid development of network technology.. For those audio and video resources, howto classify and manage these data has become an urgent problem to handle. Espec iallyfor the large number of music resources, manual classification has been unable to copewith, people need to have a fast and effective identification method for them. In the studyof music identification algorithm, the aspect of music genre classification and musicsinger recognition was carried out more widely. Among the research, the music singerrecognition usually use machine to identify and classify.The design idea of this algorithm is using the transformation of two groups ofcepstrum to reduce the interference of background accompaniment for singerrecognition.This algorithm first makes use of the speaker recognition algorithm,topre-processing and feature extraction for voice data.But different with otheralgorithms,we extract two feature vectors,one group of solo singing voice of singers,onegroup of singing voice with accompaniment.Then research the changes of the twogroups of MFCC cepstrum coefficients,and use Gauss Mixture Model to describe.At thelast,using the transformation when get the feature vector of unknown audio resources, obtainthe goal of weakening the interference of background music accompaniment,finally get thevoice information of singer,using the template matching to identify the unknown singer.Through the design and implementation of experiment, we verified the feasibility ofthis method applying to the field of singer identification, the experimental results showedthat the algorithm has some improvement on the singer recognition effect, and comparedwith other methods and their results.