Hello! I think I understand exactly the literature behind your project. I have worked on a very similar project before for real time speaker identification. The choice of mfccs is a good choice. Recently many hybrid algorithms have emerged, and I think it is a good idea to take a look on them before making the final decision.