Mel-Frequency Cepstral Coefficient (MFCC) for Music Feature Extraction for the Dancing Robot Movement Decision

摘要

In this research, we built a system that has the capability to recognize some of music patterns which each of the music pattern is obtained by cutting the whole music into some sections. The Music that used has the title of Kicir-Kicir, one of the famous traditional music belongs to Indonesia. We used the Mel Frequency Cepstral Coefficients (MFCCs) method to extract the feature of the music. The neural network (NN) is proposed as the music recognition algorithm. In this study, we used Backpropagation Neural Network (BNN) algorithm, the most popular NN and is used worldwide in many different types of applications. Each section of the music will be recorded many times to obtain the train and test set data. Those set of data will be extracted using MFCC and then classified using BNN. In the examination, one of the section of Kicir-Kicir music is recorded, then it is extracted using MFCC and recognized using BNN. Result of BNN will decide which one of movement the dancing robot should be done. Experimental results show that MFCC and BNN are capable to recognize the music with 76% of success rate.