【摘要】 为提高声品质评价模型的预测精度,提出一种从时频联合域上分析声信号特征,根据人对声音的感受特性提取相关客观参量,构建声品质评价模型的方法。采用互补集成经验模式分解对声信号进行处理,得到原信号的各阶本征模函数分量;然后通过Hilbert变换计算各本征模函数分量的瞬时频率,并引入临界频带值对瞬时频率进行计权,提取出新的客观参量—计权能量值;将计权能量值作为输入,对相关向量机模型进行训练和预测,最终得到声品质评价模型。以某混合动力汽车动力耦合机构处噪声为例,分别建立采用计权能量值的声品质评价模型和采用心理声学参数的声品质评价模型,将两种评价模型的预测结果与主观评价试验结果对比,发现前一个模型的预测结果与主观评价结果更为接近。证明从时频联合域上提取的客观参量—计权能量值能更好反映人对声信号的主观感受,声品质评价模型的预测精度也更高。更多还原

【Abstract】 In order to improve the prediction accuracy of the sound quality evaluation model, a method for analyzing the acoustic signal characteristics in the time-frequency joint domain and extracting the relevant objective parameters based on human perception of the sound characteristics is proposed. At first, the acoustic signals are decomposed by complementary integrated empirical mode decomposition to obtain the intrinsic mode function(IMF) of the original signals.Subsequently, the instantaneous frequency of each IMF is calculated by Hilbert Transform. And the instantaneous frequencies are weighted by critical frequency band values to extract the objective parameters-weighted energy values.Finally, with the parameters-weighted energy values as the input, the correlation vector machine model is trained and predicted, and the sound quality evaluation model is obtained. Taking the noise at the power coupling mechanism of a hybrid electric vehicle as an example, two acoustic quality evaluation models based on the weighted energy values and the psychoacoustic objective parameters are established respectively. The prediction results of the two evaluation models are compared with the subjective evaluation results. It is found that the prediction results of the former model are closer to the results of subjective evaluation than the latter. It is proved that the weighted energy values extracted from the time-frequency joint domain can better reflect the subjective feelings to the acoustic signals and has higher prediction accuracy.更多还原