Hiligaynon language 5-word vocabulary speech recognition using Mel frequency cepstrum coefficients and genetic algorithm

Abstract

In the study conducted by the Department of Health National Epidemiology Center, there is a high incidence and mortality rates of breast cancer among Western Visayan women, specifically in Bacolod city, Philippines. The development of breast self-examination (BSE) multimedia training system that can be easily used by the local female population in Western Visayas can help awareness and prevention of this dreaded disease. This system incorporates Hiligaynon speech recognition for motion control commands. Hiligaynon language, popularly known as Ilonggo, is an Austronesian language spoken in the Western Visayas region of the Philippines with approximately 11 million speakers, 7 million of which are native speakers. This study focuses on a 5-word vocabulary Hiligaynon language speech recognition for the BSE multimedia training system with feature extraction using Mel frequency cepstrum coefficients and pattern recognition using genetic algorithm. The genetic algorithm uses Euclidean distance, neighbourhood selection, two point crossover and elitist survival techniques. The system has an adaptive database system which improves the training and classification of the Hiligaynon words. The results showed that the combined Mel frequency cepstrum coefficients and genetic algorithm techniques used together with the adaptive database system can effectively recognized the different Hiligaynon words with 97.50% accuracy.