|Abstract||A new singer identification system is presented in this thesis. The system is based on the idea of using only the vocal segments of a song to build the model of a particular singer. The most important contribution of the technique is the way these vocal segments are located. The borders between vocal and instrumental parts are first detected with the Bayesian Information Criterion(BIC), which is fed with our new panning coefficients. Then, each segment is classified as vocal or instrumental by a decision tree based on MFCCs. Having vocal segments located, our method works like most speaker identification systems do, that is, by training a GMM for each singer through the Expectation-Maximization algorithm. The performance of the singer identification system has not been as successful as expected, however many other applications can be thought of based on the proposed segmentation technique.