Automatic Speech Recognition System using Bio-Inspired Optimization and Convolutional Neural Network

Abstract

The incorporation of machine learning methods has led to significant breakthroughs in Automatic Speech Recognition (ASR) systems. This study presents a new method that integrates the Opposition Whale Optimization algorithm (OWOA) with Convolutional Neural Network (CNN) to improve the efficiency of ASR systems. The Opposition Whale Optimization method, which draws inspiration from the social behavior of humpback whales, provides a distinct capacity to explore and exploit in order to optimize the parameters of the ASR system. The method employs opposition-based learning to achieve a well-rounded exploration of the search space, resulting in improved convergence speed and prevention of premature convergence. The use of OWOA, in combination with a CNN architecture, allows for the extraction of hierarchical characteristics from voice data. CNNs have achieved notable success in several pattern recognition applications owing to their capacity to grasp spatial and temporal connections in data. By using the capabilities of CNNs, the suggested ASR system may efficiently acquire distinctive characteristics from unprocessed audio data, resulting in enhanced accuracy in recognizing speech. The suggested technique is validated by conducting experimental assessments on typical speech datasets. The performance increases produced by integrating OWOA and CNN are assessed by conducting comparative assessments against baseline ASR systems. The findings indicate that the suggested ASR system surpasses current approaches in terms of accuracy in recognizing speech and speed of convergence, hence highlighting its potential for practical applications.