Deep Learning and Limited Boltzmann Machines for Speaker Recognition

Abstract

Speaker recognition has become an essential aspect of modern voice-based systems such as security and authentication applications. In this research, we propose a new method for speaker recognition based on deep learning and limited Boltzmann machines. The method comprises preemphasis and overlapping type framing, endpoint detection, feature extraction, and training of a depth belief network pattern using a limited Boltzmann machine layer. The Softmax graders are added in the top layer of the pattern, and the speaker's phonetic feature is input into the pattern for training. The likelihood probability of other speakers' phonetic features is calculated, and the speaker corresponding to the maximum probability is identified as the recognized result. The results show that the proposed method outperforms other state-of-the-art methods, achieving high accuracy and robustness to noise and signal variations.