| dc.description.abstract | Human auditory system is capable of extracting rich and meaningful data from complex audio signal. To recognize chord sequences that played in some kind of music is not an easy task. People need big effort to train their sense of hearing so they can recognize that kind sound of chords. This condition is also valid in a computer system. Finding the key and labeling the chords automatically from music are great use for those who want to do harmonic analysis of music. Hence automatic chord recognition has been a topic of interest in the context of Music Information Retrieval (MIR) for several years, and attempts have been made in implementing such systems using well understood signal processing and pattern recognition techniques. This research is about to recognize the sound of chord that played and recorded by guitar instrument. There are 24 major-minor chords that used in this research. MFCC is used as feature extraction and the number of coefficient cepstral that used are 13 and 26. Each chord signal that has been extracted then clustered using K-means algorithm with 8, 12, 16, 20, 24, 28, 32 k numbers to create codebook that use as a model of each chord. For the recognition process, there are two methods that used in this research, unstructured recognition and structured recognition. For the result, this research produces two kinds model of codebook that are codebook with 13 coefficients and codebook with 26 coefficients. Both types of codebook show a good result with accuracy level above 88%. The best result yielded from usage of 26 coefficient cepstral with structured recognition. It’s accuracy level reach 97%. Hence the usage of 26 coefficient cepstral is better than the usage of 13 coefficient cepstral with difference of accuration level is about 7%. This research also shows the affectation of the numbers k-means that used. An increasing accuration level shown by increasing the amount of k-cluster. | en |
