基于WPD和EMD的噪声品质预测模型

声品质作为汽车舒适性的一个重要指标,目前已经成为汽车领域一个重要的研究方向。根据人耳的听觉特性,提出一种基于小波包分解(WPD)和经验模态分解(EMD)的21个特征频带划分方法。按照所提出的方法,将采集得到的车辆噪声信号进行分解并提取信号在各频带的声能量时变特征。之后根据BP神经网络原理将提取的能量特征作为输入,计算得出响度和尖锐度等声品质评价参数作为输出,建立一种基于WPD和EMD的声品质评价模型。验证结果表明,所建立的模型可以准确地预测响度和尖锐度等心理声学参数,可作为声品质评价的一种有效方法。

Sound Quality Prediction Model of Noise Based on Wavelet Packet Decomposition and Empirical Mode Decomposition

As one of the most important vehicle comfort indices, sound quality has become an active research field. Based on the human auditory characteristics, a method with 21-feature-band allocation was presented according to wavelet packet decomposition (WPD) and Empirical Mode Decomposition (EMD). Using the proposed methods, vehicle noise signal was decomposed and its energy features of each frequency band were extracted. On the basis of BP neural network theory, taking the energy features as inputs and the loudness and sharpness calculated by using commercial software as outputs, a new model combined by the WPD and EMD for sound quality evaluation was established. The verification results show that the newly proposed model can predict the loudness and the sharpness accurately. It can be used as an effective method for sound quality evaluation.