VMD-Hilbert变换在扬声器异常声检测中的应用

针对基于时频分析的扬声器异常声检测方法中短时傅里叶变换、小波包变换存在的不足,提出了一种基于变分模态分解-希尔伯特(Variational Mode Decomposition and Hilbert,VMD-Hilbert)变换的扬声器异常声检测方法。首先通过仿真信号分析,研究了VMD-Hilbert变换的时频特性,并与其他三种时频分析进行了对比,结果表明VMD-Hilbert变换具有更好的自适应性、能量聚焦性与时频分辨率。然后,对实测扬声器声响应信号进行VMD-Hilbert变换,求得被测扬声器单元的时频矩阵与标准时频矩阵之间的特征距离,并与其它三种时频分析下的特征距离进行对比。实验结果表明,VMD-Hilbert变换下的类间特征距离的离散度较大,便于更好地设定阈值,从而验证了VMD-Hilbert变换能更好地表征异常声的时频特征,以及其在处理非线性、非平稳的扬声器声响应信号时的优越性。

The application of VMD-Hilbert transform in loudspeaker Rub & Buzz detection

In view of the shortcomings of loudspeaker Rub & Buzz detection based on time-frequency analysis, such as short-time Fourier transform and wavelet packet transform, a method of loudspeaker Rub & Buzz detection based on variational mode decomposition and Hilbert (VMD-Hilbert) transform is proposed. Firstly, the time-frequency characteristics of the VMD-Hilbert transform are studied by simulation signal analysis, and compared with the other three time-frequency analysis methods. The results show that the VMD-Hilbert transform has better adaptability, energy focus and time-frequency resolution. Then, the sound response signals of measured loudspeakers are processed with VMD-Hilbert transform to obtain the feature distances between the measured loudspeakers. The comparative analysis of feature distances obtained by different time-frequency analysis methods is made. The experimental results show that the dispersion of the feature distances between classes under VMD-Hilbert transform is larger, which is beneficial for setting the appropriate threshold. It is verified that the VMD-Hilbert transform can better represent the time-frequency characteristics of Rub & Buzz, and its superiority in dealing with nonlinear and nonstationary loudspeaker sound responses is also verified.