中文摘要：

以平板结构导纳函数为纽带,建立冲击声信号特征与声源特性之间的关联,获得与声源属性密切相关的特征用于目标分类。针对四边简支矩形被击板,借助信号参数识别算法获得与声源物理属性有关的6维导纳特征,并从冲击声样本中提取80维音色特征,将音色特征和导纳特征做相关性分析,获得与声源物理属性相关的信号特征集。利用BP神经网络进行分类,结果表明,当采用与特定声源物理属性相关的信号特征子集时,分类效果达到同组最优。

英文摘要：

Regarding the mobility of impact plates as a link, the relationships between the acoustic features and the physical properties of the sources are constructed, and more detailed descriptions of the acoustic features are acquired and more effective sound source recognitions are achieved. Aiming at a simply supported rectangular plate, through modal identification techniques, six mobility features associated with physical prop- erties are obtained. Timbre features with 80 dimensions are extracted from impact sounds. Through the correlation analysis of the acoustic features and the mobility features, a subset of acoustic feature is acquired relating to sound source properties. Then the sound source identification is conducted through the BP neural network. When using the feature subset associated with mobility features, the identification becomes best in the same group.