Automatic singer identification based on auditory features

The paper describes a method of identifying singers' voice from the monophonic music including sounds of various musical instruments based on auditory features. In this system, there are four problems to solve, vocal segment detection, feature extraction, modeling of the singing voice and identification. For a song to be identified, the vocal/nonvocal segment is detected via a new classifier—Sparse Representation-based Classification (SRC). The feature extraction is of the most importance. Human ear can distinguish among different types of sounds, so auditory features to describe the singer's voice are important. To describe the auditory features, we calculate features of each frame including Mel-frequency Cepstral Coefficient (MFCC), Liner Prediction Mel-frequency Cepstral Coefficient (LPMCC) and Gammatone Cepstral Coefficient (GTCC). Finally, we introduce the Gaussian Mixture Model (GMM) to model the singers' voice. This system is demonstrated to improve the performance of an automatic singer identification system in Music Information Retrieval (MIR).