Aiming at the two key links of loudspeaker fault diagnosis and fault recognition in the process of loudspeaker rub & buzz automatic classification, an automatic classification method of loudspeaker rub & buzz based on Variational Mode Decomposition (VMD) multiscale entropy (MSE) and Grey Wolf Optimizer-Support Vector Machines is proposed. First, the radiated acoustical signals of loudspeaker units were decomposed by VMD, and calculate the correlation coefficient of each intrinsic mode function (IMF) with the original signal. Then, select the IMF component with high correlation coefficient to extract the multi-scale entropy as the feature vector. Finally, the loudspeaker rub & buzz was judged by GWO-SVM. The experimental results show that, compared with the EMD (EMD) multi-scale entropy, VMD multiscale dispersion entropy (MDE), and EMD multiscale dispersion entropy, VMD multi-scale entropy has a higher recognition rate，The recognition accuracy rate is 99.3%.VMD multi-scale entropy can more accurately characterize the loudspeaker rub & buzz characteristics of the loudspeaker unit .