Cascaded Feedforward Neural Networks for speaker identification using Perceptual Wavelet based Cepstral Coefficients

Article type: Research Article

Authors: Renisha, G. | Jayasree, T.^{; *}

Affiliations: Department of Electronics and Communication Engineering, Government College of Engineering, Tirunelveli, Tamilnadu, India

Correspondence: [*] Corresponding author. T. Jayasre, Assistant Professor, Department of Electronics and Communication Engineering, Government College of Engineering, Tirunelveli, Tamilnadu, India. E-mail: jayasree@gcetly.ac.in.

Abstract: The rapid development in technology has led to a colossal surge in the use of biometric authentication system. Speaker identification biometric is one of the fields that is under progress and demands more and more precision. The objective of this research is to explore the issue of identifying a speaker from voice regardless of the content. Perceptual Wavelet Packet Transform (PWPT) and Artificial Neural Networks (ANN) approach are discussed in this paper for speaker identification. Perceptual Wavelet Packet Cepstral Coefficients (PWPCC) are used for transforming speech into spectral feature vectors, and the most germane aspects of the speech signal are selected from the energy and variance distribution characteristics. These selected attributes are presented to the Cascaded Feedforward Neural Network (CFNN) and trained with Levenberg-Marquardt Back Propagation (LMBP) algorithm for further classification. The performance of the network is determined by evaluating the Speaker Identification Rate (SIR). For comparison, five different gradient descent training algorithms are considered and it is found that the LMBP produces better performance. The proposed model is evaluated for clean as well as noisy speech at various SNR levels and is found to be competitive, and the experimental results show significant improvement in speaker identification rate compared with other classical methods.

Keywords: Perception, wavelet, speaker, speech, neural network

DOI: 10.3233/JIFS-182599

Journal: Journal of Intelligent & Fuzzy Systems, vol. 37, no. 1, pp. 1141-1153, 2019