Automatic Isolated Speech Recognition System Using MFCC Analysis and Artificial Neural Network Classifier: Feasible For Diversity of Speech Applications

In this research, an Automatic Isolated Speech Recognition System (AISR) has been proposed to categorize four distinct word class pronounced as "GO"; "STOP"; "LEFT" and "RIGHT". Speech data has investigated mainly employing Mel Frequency Cestrum analysis together with specific subsidiary discrete signal processing to generate compact and productive speech database. In this paper, an Artificial Neural Network (ANN) has been used as a classifier to categorize four distinct word classes. The ANN functions with the Feed Forward Multi-layer Perceptron (FFMP) with Back Propagation (BP) training algorithm. Trial and error procedure is adopted to determine an optimally performed ANN classifier. The presented method has achieved the best recognition rates 97.14% and 98.57% respectively for two training algorithms- Steepest Gradient Descent with Adaptive Learning and Scaled Conjugate Gradient by investigation on different configuration and parameters of the ANN. To validate the feasibility and efficiency of the proposed ASR system, a small-scale application has implemented where two filament bulbs have been efficiently operated with designated speech classes.