Features for Audio Classification

Four audio feature sets are evaluated in their ability to differentiate five audio classes: popular music, classical music, speech, noise and crowd noise. The feature sets include low-level signal properties, mel-frequency spectral coefficients, and two new sets based on perceptual models of hearing. The temporal behavior of the features is analyzed and parameterized and these parameters are included as additional features. Using a standard Gaussian framework for classification, results show that the temporal behavior of features is important for automatic audio classification. In addition, classification is better, on average, if based on features from models of auditory perception rather than on standard features.