Speaker identification based on spectrogram and local binary patterns

This paper presents a text-independent, closed-set speaker identification approach based on spectrogram and dynamic time warping (DTW) algorithm. The preprocessed speech signals are divided into some chunks, then calculated to get the magnitude of the frequency spectrum, which creates the spectrograms. The local binary patterns (LBP) operator are used to obtain the LBP vectors being treated as the speech features. The distances between each of the LBP vectors are measured by DTW algorithm, which aims to align two sequences of input LBP vectors by warping the time axis iteratively until an optimal match between the two LBP vectors is found. Through this elastic and robust sequential data matching, the proposed method identifies which one is the target speaker among a closed-set of speakers. The numerical experiments are carried out to verify the theoretical results and clearly show that our identification method has an acceptable accuracy. ©, 2015, Binary Information Press. All right reserved.