Direction of arrival estimation of an acoustic wave using a single structural vibration sensor
Direction of arrival estimation of an acoustic wave using a single structural vibration sensor
复制标题
使用单个结构振动传感器估计声波的到达方向
DOI：
10.1016/j.jsv.2023.117671
复制
发表时间：
2023
影响因子：
4.7
通讯作者：
Bocko, Mark F.
中科院分区：
文献类型：
--
作者：
DiPassio, Tre;Heilemann, Michael C.;Bocko, Mark F.
柔性板对入射声压波的模态振动响应与声压波的到达方向有关。这项工作提出了一个概念证明，即振动测量的平板结构传感器被用来推断入射声波的波达方向。在这里报告的实验中，具有各种材料阻尼因子的矩形面板在消声空间中被以水平面中的-90 °和+90 °之间的角度入射的声波激励。梅尔频率倒谱系数（MFCC）从记录的面板响应计算，并用于训练深度神经网络（DNN）来估计DOA。实验结果表明，在受控条件下，利用单个结构传感器的记录，可将含宽带噪声的入射声波的DOA估计到±5°以内，可靠性达99.8%。对于包含频谱不太均匀的人类语音信号的声波，使用来自单个传感器的数据训练的DNN在实验条件下正确估计DOA，其可靠性为86.5%，在±5°内，可靠性为96%，在±10°内。这项工作的范围是建议，一个面板的模态响应可能包含足够的信息，使DOA估计少到一个传感器。因此，提出了一个概念的证明，而不是一个最终的工程解决方案，沿着的实验限制和未来的改进讨论。
The modal vibrational response of a flexible panel to an incident acoustic pressure wave is dependent on the direction of arrival (DOA) of the wave. This work presents a proof of concept whereby vibration measurements of flat panels made by structural sensors were used to infer the DOA of an incident acoustic wave. In the experiments reported here, rectangular panels with various material damping factors were excited in an anechoic space by acoustic waves incident at angles between− 90° and+ 90° in the horizontal plane. Mel-frequency cepstral coefficients (MFCCs) were computed from the recorded panel responses and used to train a deep neural network (DNN) to estimate the DOA. Experimental results show that under controlled conditions, the DOA of incident acoustic waves containing broadband noise may be estimated to within±5° with a reliability of 99.8% utilizing recordings from a single structural sensor. For acoustic waves containing less spectrally uniform human speech signals, a DNN trained using data from a single sensor correctly estimated DOA to within±5° with a reliability of 86.5% and to within±10° with a reliability of 96% within the experimental conditions. The scope of this work is to suggest that a panel’s modal response may contain sufficient information to enable DOA estimation with as few as one sensor. As such, a proof of concept is presented in place of a final engineering solution, along with a discussion of experimental limitations and future improvements.
使用两个正交一阶差分麦克风阵列进行深度学习辅助声源定位。
DOI：
10.1121/10.0003445
发表时间：
2021
期刊：
The Journal of the Acoustical Society of America
影响因子：
--
作者：
Nian Liu;Huawei Chen;Kunkun SongGong;Yanwen Li
通讯作者：
Yanwen Li
使用振动面板表面上的结构传感器进行音频捕获
影响因子：
1.4
作者：
Dipassio, Tre;Heilemann, Michael C.;Bocko, Mark F.
通讯作者：
Bocko, Mark F.
影响因子：
4.7
作者：
MITCHELL, AK;HAZELL, CR
通讯作者：
HAZELL, CR