Sistem Penilaian Berdasarkan Tepuk Tangan Menggunakan MFCC dan Codebook

Abstract

Scoring system based on sound of applause is often used in many contests in Indonesia. However, determining the winner in a conventional way tend to be subjective. This research develops automated computer-based scoring system to count the number of people who applauded and determine the winner of a competition based on applause. This research proposes Mel Frequency Cepstral Coefficient (MFCC) as a method of feature extraction, while Codebook as the pattern recognition. They produce a model in the form of classes that are clustered by K-Means method. The important parameter in MFCC and Codebook are the number of cepstral coefficients, overlap, the time frame, and the number of clusters. The system implements optimum parameters with maximum accuracy of 83.3%. However the accuracy of the real system is only 50%, lower than the accuracy obtained by using a simpler threshold, which is %. Thus, this study shows that the frequency-based MFCC method is not suitable to be applied for this application. Rather, it is recommended to use an amplitude-based method