Music emotion classification using double-layer support vector machines

This paper presents a two-layer system for detecting emotion in music. The selected target emotion classes are angry, happy, sad, and peaceful. We presented an audio feature set comprising the following types of audio features: dynamics, rhythm, timbre, pitch, and tonality. With the feature set, a support vector machines (SVMs) is applied to each target emotion class with calm emotion as the background class to train a hyperplane. With the four hyperplanes trained from angry, happy, sad, and peaceful, each test clip can output four decision values. This decision values are regarded as the new features to train a second-layer SVMs for classifying the four target emotion classes. The experiment result shows that our double layer system has a good performance on music emotion classification.