Klasifikasi genre musik menggunakan Learning Vector Quantization (LVQ)

Abstract

Radio stations and music television have a milion of music tapes. A lot of musical genres create a problem when people wants to determine the right genre of a new kind of music. To classify the musical genre is not an easy task, because the musical genre is really difficult to standardization. Automatic musical genre classification can assist the human role in that process and help people to searching for the song acording to the genre that people want. This research using Mel Frequency Coefficient Cepstrum (MFCC) to obtain feature extraction. Learning Vector Quantization (LVQ), one kind of artificial neural network used for classification method. The number of genres that are used were four kind of musical genre, that is rock, classic, keroncong, and jazz with four different duration that is 5 second, 10 second, 20 second and 25 second. This research using k- fold cross validation to distribute dataset for training and testing set with the number of folds as much as 2 until 10 fold. This research succesfully implemented MFCC feature exraction and classification using LVQ. Based on this research, the accuracy of the classification using Learning Vector Quantization reaches 93,75% for the four type musical genre. The highest accuracy value was obtain from the experiments with a duration of 10 second and the number of fold 4. Training time for each duration is 30 minute for 5 second music duration, 45 minute for 10 second music duration 120 minute for 20 second music duration and 150 minute for 25 second music duration.