Wall pressure fluctuations generated by Turbulent Boundary Layers(TBL) provide a significant contribution in reducing the structural vibration and the aircraft cabin noise. However,it is difficult to evaluate these fl...Wall pressure fluctuations generated by Turbulent Boundary Layers(TBL) provide a significant contribution in reducing the structural vibration and the aircraft cabin noise. However,it is difficult to evaluate these fluctuations accurately through a wind tunnel test because of the pollution caused by the background noise generated by the jet or the valve of the wind tunnel. In this study, a new technology named Subsection Approaching Method(SAM) is proposed to separate the wall pressure fluctuations from the background noise induced by the jet or the valve for a transonic wind tunnel test. The SAM demonstrates good performance on separating the background noise from the total pressure compared to the other method in this study. The investigation considers the effects of the sound intensity and the decay factor on the sound-source separation. The results show that the SAM can derive wall pressure fluctuations effectively even when the level of background noise is considerably higher than the level of the wall pressure fluctuations caused by the TBL. In addition, the computational precision is also analyzed based on the broad band noise tested in the wind tunnel. Two methods to improve the precision of the computation with the SAM are also suggested: decreasing the loop gain and increasing the sensors for the signal analysis.展开更多
文章提出了一种基于深度神经网络的低延迟(≤算法延迟20 ms)声源分离方法。方法利用了扩展的过去的上下文,输出软时频掩码用于分离音频信号,比基本的NMF有更好的分离性能。实验表明,基于DNN的方法比起基本的低延迟的NMF方法在不同帧长...文章提出了一种基于深度神经网络的低延迟(≤算法延迟20 ms)声源分离方法。方法利用了扩展的过去的上下文,输出软时频掩码用于分离音频信号,比基本的NMF有更好的分离性能。实验表明,基于DNN的方法比起基本的低延迟的NMF方法在不同帧长的处理帧和分析帧上,SDR平均至少提升1 d B,尤其是当处理帧较短时,效果尤为显著。展开更多
文摘Wall pressure fluctuations generated by Turbulent Boundary Layers(TBL) provide a significant contribution in reducing the structural vibration and the aircraft cabin noise. However,it is difficult to evaluate these fluctuations accurately through a wind tunnel test because of the pollution caused by the background noise generated by the jet or the valve of the wind tunnel. In this study, a new technology named Subsection Approaching Method(SAM) is proposed to separate the wall pressure fluctuations from the background noise induced by the jet or the valve for a transonic wind tunnel test. The SAM demonstrates good performance on separating the background noise from the total pressure compared to the other method in this study. The investigation considers the effects of the sound intensity and the decay factor on the sound-source separation. The results show that the SAM can derive wall pressure fluctuations effectively even when the level of background noise is considerably higher than the level of the wall pressure fluctuations caused by the TBL. In addition, the computational precision is also analyzed based on the broad band noise tested in the wind tunnel. Two methods to improve the precision of the computation with the SAM are also suggested: decreasing the loop gain and increasing the sensors for the signal analysis.
文摘文章提出了一种基于深度神经网络的低延迟(≤算法延迟20 ms)声源分离方法。方法利用了扩展的过去的上下文,输出软时频掩码用于分离音频信号,比基本的NMF有更好的分离性能。实验表明,基于DNN的方法比起基本的低延迟的NMF方法在不同帧长的处理帧和分析帧上,SDR平均至少提升1 d B,尤其是当处理帧较短时,效果尤为显著。
