Speaker verification deals with the task of confirming the identity of a claim using a hypothesized speaker model and a speaker model database. This work concentrates on a speaker verification system by combining GMM and SVM. The feature vectors used for modelling are Mel Frequency Cepstral Coefficients (MFCC). The database is collected through different recording equipments which is considered as different channels. In order to reduce the channel effects, a method called feature mapping is implemented. To model a speaker in SVM positive as well as negative files are needed. Comparative studies on the rate of modelling using different types of negative files are performed in this work. Then the proposed system is tested over the Malayalam database. The maximum modelling rate obtained is in the range of 95–99%. Then the accuracy of the system with different type of negative files are calculated and compared the results. The system is developed using MATLAB 7.12.0(R2011a) and LIBSVM tool kit.