飞机驾驶舱话音记录器(CVR)记录的舱音信号,通常是由语音、警告声、开关按钮声和背景噪声等混合而成的。目前对该类信号的分析和辨别主要是通过CVR译码专业设备下载后依靠人耳辨听,存在不易准确分辨出各种独立声音信号的缺点。应用基于独立分量分析(ICA)方法的快速ICA(FastICA)、自然梯度、JADE算法,以及扩展优化的COMBI等算法,对舱音混合信号进行分离,并对各种算法的分离效果进行比较。仿真结果表明,上述算法可以有效地将CVR混合信号中的独立声音信号分离出来,其中COMBI具有更为优越的分离效果。

The acoustic signal recorded by Cockpit Voice Recorder (CVR) on airplanes is a mixed signal composed of voices,alarm sound and noises. So far,the analysis and identification of these acoustic signals are mainly depended on human audition,which is difficult to separate independent sounds. Algorithms based on Independent Component Analysis(ICA), such as FastICA,natural gradient algorithm,Joint Approximate Diagonalization of Eigenmatrix (JADE) algorithm and optimized COMBI (Combine of WASOBI and EFICA) etc. are used to separate specific sound from acoustic signal effectively. Comparisons of simulation results using these different algorithms are made,which shows that the COMBI algorithm has a better performance in separating independent sounds from acoustic signal.