Improving automatic music genre classification with hybrid content-based feature vectors

Abstract

Current research on the task of automatic music genre classification has been focusing on new classification approaches based on combining information from other sources than the music signal. The reason for this is that the use of contentbased approaches, i.e. using features extracted directly from the audio signal, seems to have reached a glass ceiling. In this work we show that by using different types of contentbased features together it is possible to substantially improve the classification accuracy. This is an interesting result as different types of content-based features aim, at a conceptual level, to capture the same type of information. In order to identify which types of content-based features are responsible for the predictive accuracy gain, we also used a feature selection (FS) approach based on a genetic algorithm (GA). The analysis of the results in two databases shows that the use of the GA for FS succeeds in selecting a representative subset without significant loss in accuracy. It also shows that all the different types of content-based features employed are important for the improvement of the accuracy in classifying music genres.

A Survey of Evaluation in Music Genre Recognition

“…We see that the most common experimental design to test MGR systems is Classify. More than 91% (397) 2 of the referenced work having an experimental component (435) uses such a design [1-9, 11- 21, 26-43, 45,47-50,52-56,58-60,62-65,68-70,72-76,79-83,86-90,92-99,102-106,108-122, 124-135, 137-148, 150, 151, 153-165, 167-172, 174-180, 182-197, 199-202, 204- [271] test a neural network trained to discriminate between classical and pop music. They extract features from the audio, input them to the neural network, and compare the output labels against those they assigned to the excerpts of their dataset.…”

An Evaluation of Symbolic Feature Sets and Their Combination for Music Genre Classification

“…We also plan to extend this study to investigate the use of content-based symbolic features for hierarchical music genre classification [28]. Another interesting research direction is to compare the performance of the early fusion method used in this work with the late fusion employed by several authors in previous studies [11], [14], [25], [29]. We would like to thank the authors of [25] for kindly sharing the information about the song artist and song title used in their three-fold cross-validation experiments.…”

“…One of the few exceptions is the work of [14], where the authors combine different types of content-based audio features. The combination of different types of audio features improved the music genre classification accuracy [14].…”

“…One of the few exceptions is the work of [14], where the authors combine different types of content-based audio features. The combination of different types of audio features improved the music genre classification accuracy [14]. For this reason, in this study we evaluate whether it is possible to improve the music genre classification accuracy by combining two different types of content-based symbolic features, which to the best of the authors knowledge has not been done before.…”

Music genre classification using On-line Dictionary Learning

“…Costa et al further improved upon their method in [8], where LBP features were combined with Mel Scale Zoning (MSZ) technique [9] to obtain better classification on the LMD database using the SVM classifier. Gradually, music genre classification moved in the direction of hybrid content based classifiers with Rhythm Histograms (RH), InsetOnset Interval Histogram Coefficients (IOIHC) and Statistical Spectrum Descriptors (SSD) being used in [10] along with MARSYAS features. Upon application of SVM for classification, the hybrid classifiers outperformed the methods involving single feature sets.…”