We analyze in this paper inherent limitations of prediction systems for speaker identification. We introduce different approaches for enhancing these systems and present results of tests on TIMIT with neural net predictive systems.
Keywords: speaker recognition, predictive neural networks, sequence classification