关键词: 多尺度时频调制, 多线性主成分分析, 听觉谱图, 支持向量机, 乐器识别

Abstract: Aiming at time or frequency feature, cepstrum feature, sparse feature and probability feature's poor classification performance for kindred and percussion instrument, an enhanced model for extracting time-frequency information and with lower redundancy was proposed. Firstly, a cochlea model was set to filter music signal, whose output was called Auditory Spectrum (AS) containing harmonic information and close to human perception. Secondly, time-frequency feature was acquired by Multiscale Time-Frequency Modulation (MTFM). Then, dimension reduction was implied by Multilinear Principal Component Analysis (MPCA) to preserve the structure and intrinsic correlation. Finally, classification was conducted using Support Vector Machine (SVM). The experimental results show that MTFM's average accuracy is 92.74% on IOWA database and error rate of percussion or kindred instrument is 3% and 9.12%, which wins out the features mentioned above. The accuracy of MPCA was higher 6.43% than that of Principle Component Analysis (PCA). It is proved that the proposed model is an option for kindred and percussion instrument identification.

Key words: Multiscale Time-Frequency Modulation (MTFM), Multilinear Principal Component Analysis (MPCA), Auditory Spectrum (AS), Support Vector Machine (SVM), musical instrument identification