Feature extraction of the Indonesian phonemes using discrete wavelet and wavelet packet transform

Abstract

The development of the pattern recognition techniques in voice recognition has rapidly increased. Many methods are used to create a system that is reliable and easily accessible so that the recognition accuracy is high. To find the pattern recognition system that is robust and reliable, proper feature extraction methods are needed. In this case, the proper method of the feature extraction is an efficient and effective method when it is used in recognizing specific features of data. This paper presents an effective and efficient method of the extracting features for speech processing. The focus of this work is to explore the feature extraction methods which are the most efficient and effective in recognizing Indonesian phonemes. The Feature extraction methods used and compared in the study are the Discrete Wavelet Transform (DWT) and Wavelet Packet Transform (WPT). The wavelet transform was done on the 2nd level until 4th level of decomposition. The comparison of the performance of both feature extraction methods are presented at the end of this section, with a statistical comparison, to draw a conclusion that the DWT methods have a better performance in terms of effectiveness and efficiency compared with the WPT method. The results of the study show that the effectiveness ratio is 60% versus 40% and the efficiency ratio is 57% versus 43%.