Voice recognition enhancement by genetic algorithm

Abstract

This research paper investigates the application of Genetic Algorithms (GA) in optimizing Artificial Neural Networks (ANN) for phoneme recognition. The study examines the formalism of GA, their parameters, and operators, and describes the genetic strategy adopted for phoneme recognition using the TIMIT sound database. The paper presents the outcomes of experiments conducted on the phonemes of the test base and the DR1 dialect learning base of the TIMIT base, and compares the recognition rates obtained by learning and testing with those guided by Self-Organizing Map (SOM) experiments. The findings suggest that the GA algorithm provides improved recognition rates and extends the search space to a global optimum, but can sometimes produce unacceptable results due to rapid and premature convergence and the overfitting problem. The study proposes a hybrid model between SOM and GA to take advantage of each of their properties and improve recognition rates. The results show that the proposed approach achieves an average recognition rate of over 95%.