Bacterial Foraging Optimized Parameters for ANN using Adaptive Harris Hawks Weight Optimization

Abstract

Speaker recognition is important to validate user identity using the extracted features of the audio speech signal in the field of authentication and surveillance. Two modules may be used to understand the speaker, namely training and testing. The capability of recognition systems to identify speakers based on waveform distribution depends largely on how the recognition system trains model parameters to provide the best class of discrimination. The mel-frequency cepstral coefficients (MFCCs) of each speaking sample are obtained initially in the training phase by preprocessing the audio speech signal. The characteristics are then identified using RBF-ANN to the target speaker. Recognition is based on an estimation of a sufficiently large number of acoustic features. In the proposed work, Bacterial Foraging Optimized (BFO) parameters are used that are provided as input for the RBF-ANN model. The ANN weights are updated using the Adaptive Harris Hawks Optimization (AHHO) method for improving the system performance. The performance of the proposed DNN-RBF based AHHO is compared with three different deep learning based optimization algorithms Modified Grey Wolf Optimization (MGWO), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO) and the results show that the proposed model accuracy in speaker recognition is high when compared to traditional methods.