Toward polyphonic musical instrument identification using example-based sparse representation

Musical instrument identification is one of the major topics in music signal processing. In this paper, we propose a musical instrument identification method based on sparse representation for polyphonic sounds. Such the identification has been still categorized into challenging tasks, since it needs high-performance signal processing techniques. The proposed scheme can be applied without any signal processing such as source separation. Sample feature vectors for various musical instruments are used for the base matrix of sparse representation. We conducted two experiments to evaluate the proposed method. First, the musical instrument identification is tested for monophonic sounds using five musical instruments. The average accuracy of 91.9% was obtained and it shows the effectiveness of the proposed method. Second, musical instrument composition of polyphonic sounds is examined, which contain two instruments. It is found that the estimated weight vector by sparse representation indicates the mixture ratio of two instruments.