A Machine Learning Approach to Automatic Music Genre Classification

Carlos N Silla,Alessandro L Koerich,Celso A A Kaestner

doi:10.1007/bf03192561

Abstract

Abstract This paper presents a non-conventional approach for the automatic music genre classification problem. The proposed approach uses multiple feature vectors and a pattern recognition ensemble approach, according to space and time decomposition schemes. Despite being music genre classification a multi-class problem, we accomplish the task using a set of binary classifiers, whose results are merged in order to produce the final music genre label (space decomposition). Music segments are also decomposed according to time segments obtained from the beginning, middle and end parts of the original music signal (time-decomposition). The final classification is obtained from the set of individual results, according to a combination procedure. Classical machine learning algorithms such as Naïve-Bayes, Decision Trees, k Nearest-Neighbors, Support Vector Machines and Multi-Layer Perceptron Neural Nets are employed. Experiments were carried out on a novel dataset called Latin Music Database, which contains 3,160 music pieces categorized in 10 musical genres. Experimental results show that the proposed ensemble approach produces better results than the ones obtained from global and individual segment classifiers in most cases. Some experiments related to feature selection were also conducted, using the genetic algorithm paradigm. They show that the most important features for the classification task vary according to their origin in the music signal.

Highlights

Music is nowadays a significant part of the Internet content: the net is probably the most important source of music pieces, with several sites dedicated to spreading, distributing and commercializing music
Wiggins and Crawford [5] raise this question and argue that genre labeling is affected by two major factors: the first one is related to how the composer intended to draw upon different stylistic elements from one or more music genres; the second one is related to the social and cultural background of any participant involved in labeling the database
The following analysis can be done on the results shown in Table 2: (a) for the J48 classifier the RR approach increases accuracy, but for the OAA approach the increment is not significative; (b) the 3-NN classifier presents the same results for every segment, but they vary according to the adopted strategy; (c) for the Multi Layer Perceptron neural network (MLP) classifier the decomposition strategies increase accuracy in the beginning segment for RR, and in the end segment for OAA and RR; (d) for the Naïve-Bayes classifier (NB) classifier the approaches did not increase significantly the classification performance; and (e) the Support Vector Machines (SVM) classifier presents the best classification results, when using RR, and the worst one in every segment when employing the OAA approach

Summary

Introduction

Music is nowadays a significant part of the Internet content: the net is probably the most important source of music pieces, with several sites dedicated to spreading, distributing and commercializing music. In this context, automatic procedures capable of dealing with large amounts of music in digital formats are imperative, and Music Information Retrieval (MIR) has become an important research area. The music genre is a descriptor that is largely used to organize collections of digital music It is a crucial metadata in large music databases and electronic music distribution (EMD) [1], and the most frequent item used in search queries, as pointed by several authors [8], [20], [30], [35], [36]

Methods

Results

Conclusion