Back-Propagation Neural Network Based Speaker Identification Under Noise Distortion

Abstract

The aim of this paper is to evaluate the performance of the proposed speaker identification system where Wiener filtering technique has been used to eliminate the background white Gaussian noises and linear discriminant analysis has been used to reduce the dimension of the speech features. Since the audio signal captures more environmental noises than other biometrics modalities, the main emphasis of this paper is to analyze the problem domains of noise robust speaker identification which will further improve the performance of the system. Experimental results show the superiority of the proposed system over existing methodologies of speaker identification. NOIZEUS speech database has been used to measure the performance of the proposed system where eight different noisy environmental speeches are considered such as airport, babble, car, exhibition, restaurant, street, train and train station with four different SNRs i.e., 0, 5, 10 and 15 dB. Finally the proposed system performance has been compared with existing principal component analysis based speaker identification system and shows the superiority of the proposed system.

Keywords: Speaker identification, speech with noise distortion, linear discriminant analysis, speech parameterization, back-propagation neural network

Cite this Article

Md. Rabiul Islam. Back-Propagation Neural Network Based Speaker Identification Under Noise Distortion. Trends in Electrical Engineering (TEE). 2015; 5(2): 1–6p.