DNN-RBF & AHHO for Speaker Recognition using MFCC
Speaker Recognition is essential in the field of authentication, and surveillance to validate the user’s identity using extracted feature characteristics of audio speech signal. In this work, the speaker recognitions performed by deep neural network-Radial Basis Function (DNN-RBF). Initially, the available speech signals are preprocessed to remove the noise from the input signal. The noise removal in the input signal is performed by wiener filter. From this pre-processed signal Mel frequency cepstral coefficients (MFCC) features are extracted. The i-vector is estimated from the Gaussian Mixture Model (GMM) super vector in which the dimensionality of extracted features is reduced.Extracted i-vector features are then injected within classifier for recognizing the specific speaker. Based on these extracted features, the speakers are recognized by Adaptive Harris Hawk Optimization (AHHO) based DNN-RBFin an appropriate manner. The performance of this speaker recognition process is evaluated with TIMIT (Texas Instruments/Massachusetts Institute of Technology)dataset. Some of the performance metrics like precision, accuracy, and recall are evaluated to evaluate the effectiveness of this proposed technique.The proposed speaker recognition technique is evaluated with various performance measures such as EER, precision, recall, and accuracy. The accuracy, precision, and recall values attained by proposed AHHO based DNN-RBF is 94.92%, 89.87 and 94.67 respectively. The presence of adaptive optimization approach improves the performance of DNN-RBF in speaker recognition. The implementation process is performed in Mat lab platform.
Full Text Attachment