摘要
提出了基于自相关原理提取音调的具体方法和步骤,以乐音和舰船辐射噪声为例,采用3层BP神经网络目标分类器进行仿真实验。仿真结果证明音调是有效的识别特征,且音调的识别率与目标辐射声信号的频率特性的可分性有关。频率特性的可分性越高,则目标识别率越高;当可分性不高时,可以将音调和其他主观量结合共同作为识别特征。实验表明,将特性响度作为目标特征,与音调特征结合起来,目标识别率得到进一步提高。
An autocorrelation-based algorithm is intxoduced to extract the pitch. With the pitch of music and radiated noise from ships, target classification experiments using neural network are carried out. It has been found that pitch is an effective target feature. Recognition accuracy mostly depends on the frequency difference among different targets： the larger the difference, the higher the recognition accuracy. When the frequency difference is not large enough, it is necessary to combine pitch with other psychoacoustic parameters. Simulations show that recognition accuracy can be improved by combining pitch with specific loudness mainly corresponding to the magnitude characteristics for a given sound stimuli.
出处
《声学技术》
CSCD
北大核心
2007年第4期669-673,共5页
Technical Acoustics
关键词
音调
基频
自相关
特性响度
目标识别
pitch
fundamental frequency
autocorrelation
specific loudness
target recognition
