M-SVMS and their application to output-based objective music articulation assessment

This paper proposes an approach to output-based objective music articulation assessment using M-SVMs (Multi-category support Vector Machines). This method extract MFSC (Mel Frequency Spectral Coefficient) as features, then mapping from MFSC vectors to subjective articulation evaluation levels by M-SVMs. The output results are objective music articulation assessment levels. Besides, we compare it with the situation that extracts MFCC (Mel Frequency Cesptral Coefficient) as characteristic parameters. Experimental results show that the MFSC is better and the correlation degree reaches over 0.96 when using the train samples and attains more than 0.82 when using the test samples.