Abstract
Speaker Recognition is considered as one of the primary tasks in speech processing. Nowadays, the speaker identification method has been extensively appealing for its broad application in many fields, such as smart environments, securing the cyber-physical systems, speech communications, and robotic controls. Researchers are targeting to perform an effective method that makes it possible to obtain the recognition ability that is close to the hearing of human. In order to get high accuracy, challenges of large-scale applications of speaker identification are overcome through applying techniques not only traditional models based on the GMM, but also deep learning methods. Aiming at effectively dealing with this challenge, in this paper, we present a novel model to increase the recognition accuracy of the short utterance speaker recognition system. We developed a technique to train a Neural Network (NN) on the extracted Mel-Frequency Cepstral Coefficient (MFCC) features from audio samples. Therefore, the recognition system gains the significant accuracy. The model was trained using open-source high-level neural networks API Keras.