The Effect of Audio Snippet Locations and Durations on Genre Classification Accuracy Using SVM

Abstract: Real world scenarios where machine learning based music genre classification could be applied includes; streaming services, music distribution platforms and automatic tagging of music libraries. Music genre classification is inherently a subjective task; there are no exact boundaries that separate different genres. Machine learning based audio classification attempts to classify audio by comparing feature vectors. Which features to extract from which parts of the audio greatly impact the classification accuracy. This paper investigates whether different audio snippet locations and durations impact the classification accuracy. A number of experiments were run across six genres, four kinds of snippet locations and eight durations. The results show that these parameters do in fact have a significant impact on the accuracy.