Voice authentication is a fundamental topic of research in today’s technology. Reliable speech recognition is hard to achieve, but many approaches have been proposed in recent years to achieve such with an improved degree of accuracy. The following paper presents a novel approach through which users can be authenticated with reasonable accuracy using a small voice sample. The proposed method uses MFCCs, a well known methodology for extracting features from the voice sample and finally uses Gaussian Mixture Models (GMM) for classification. An advantage of using MFCCs as the speech features is that the model is language independent. A model trained in one language can work equally well for a model trained in a different language.