摘要
针对声品质评价过程中线性回归模型评价结果的不足,采用BP神经网络对人的主观评价结果进行预测。采集摩托车在不同发动机转速下驾驶员耳旁的声信号样本,采用分组成对比较法进行主观评价试验,选取了响度、尖锐度、粗糙度作为神经网络模型输入参数,结合主观评价结果对模型进行训练与检验,并与线性回归模型输出结果进行比较。结果表明,选取驾驶员双耳响度、尖锐度、粗糙度作为模型输入能够较为准确地反映人耳对摩托车噪声的主观感觉。
BP neural networks method was selected to predict the subject evaluation results,which is more precise than the linear regression method in the process of sound quality evaluation. The noise samples near the motorcycle drivers' ear position at various engine speeds were selected, with subjective evaluation testing carried on by the way of grouped pair-wise comparison. BP neural networks model was trained, while loudness, sharpness and roughness were selected as input parameters. And the sound quality evaluation results were investigated in eom- parlson with that of the linear regression model. It reveals that the human subjective feelings to motorcycle noises could be in good prediction through neural networks method when loudness, sharpness and roughness were selected as input parameters.
出处
《世界科技研究与发展》
CSCD
2012年第5期784-786,820,共4页
World Sci-Tech R&D
基金
国家自然基金(50975296)
重庆市科技计划攻关(CSTC
2009AC6076
CSTC
2009AC6206)资助
关键词
声品质
神经网络
主观评价
心理声学参数
sound quality
neural network
subjective evaluation
psychoacoustic parameter