Natural language processing: Speaker, language, and gender identification with LSTM

Long short-term memory (LSTM) is a state-of-the-art network used for different tasks related to natural language processing (NLP), pattern recognition, and classification. It has been successfully used for speech recognition and speaker identification as well. The amount of training data and the ratio of training to test data are still the key factors for achieving good results, but have their implications on the real usage. The main contribution of this paper is to achieve a high rate of speaker recognition for text-independent continuous speech using small ratio of training to test data, by applying long short-term memory recursive neural network. A comparison with the probabilistic feed-forward neural network has been made for speaker recognition as well as gender and language identification.