The Development Speech Recognition Model using MFCC as Feature Extraction and PNN as Pattern Recognition

Abstract

Utilizing human voice to develop advanced technology is very popular at this modern days; it has many different functions. The objective of this research is to develop a recognition pattern model for words that are usually used by operation system. Eventually it can be utilized to control the operation system by voice. The chosen words to build the recognition model are 20 words: cancel, close, copy, cut, delete, edit, format, insert, maximize, minimize, move, new, open, paste, print refresh, rename, restart, search, and view. In order to get the model for each pattern, 50 times soundwave samples per word were taken. The waveforms are first recorded by computer and saved into the wave file format. The silent part are deleted, standardization are performed followed by feature extraction. The results of feature extraction process are divided into two set of data, the training data and testing data. The ratio of training data to testing data are 75%:25%, 50%:50%, and 25%:75%.The pattern recognition model is built using PNN. Experiment are then performed to measure the performance of the model. For data percentage 75% training data and 25% testing data the general accuracy result is 86%, the result that used 50% training data and 50% testing data is 87%. The last testing result that used 25% training data and 75% testing data is 88%.