Analysis of amplitude threshold on speech recognition system

Development of speech recognition systems continues to be carried out by many researchers. In many researches, system recognition accuracy is still as a main point which need to be improved. In addition to accuracy, systems algorithms computational time also becomes an important point that must be considered in developing a speech recognition system. This paper carries out a research on an analysis of initial processing stages in a speech recognition system. The initial processing stage of a speech recognition system is filtering which includes threshold analysis of filter and number of speech signal indicator data cuts. Research was carried out by testing range values of threshold and speech signal data cuts as well as observing effect of speech recognition systems accuracy. This research employed Mel Frequency Cepstral Coefficients (MFCC) as a feature extraction method, while the Euclidean distance method was used for classification. Results show that threshold values and number of speech signal data cuts affect speech recognition systems accuracy level. The highest speech recognition system accuracy is of 90 and is achieved at threshold value of 0.025, and of 3600 data cuts length. In addition, computational time of speech recognition system algorithm also influences speech signal data numbers used in computing process.