Page 145 - 《应用声学》2021年第1期

第 40 卷 第 1 期 李悦等： 非高斯环境下的深度学习脉冲信号去噪与重构 141

[5] Meng Z, Li J, Zhao Y, et al. Conditional teacher-student
5 结论 learning[C]//IEEE International Conference on Acous-
tics, Speech and Signal Processing (ICASSP), 2019.
本文提出一种基于深度学习的能够在高斯和 [6] Zhang K, Zuo W, Chen Y, et al. Beyond a Gaussian
denoiser: residual learning of deep CNN for image de-
非高斯环境噪声中增强与重构水声脉冲信号的方
noising[J]. IEEE Transactions on Image Processing, 2017,
法。通过建立全卷积的脉冲信号去噪模型，学习接 26(7): 3142–3155.
收信号 STFT特征与接收信号和纯净信号 STFT特 [7] Guo S, Yan Z, Zhang K, et al. Toward convolutional blind
征残差之间的映射，实现噪声抑制，最后通过逆变换 denoising of real photographs[C]//IEEE/CVF Conference
on Computer Vision and Pattern Recognition (CVPR),
重构脉冲信号。仿真实验结果显示，模型可在 3 种 2019.
背景噪声中有效实现脉冲信号的去噪与重构，相较 [8] 孟庆松, 王彬, 邵高平. α 稳定分布噪声下水声线性调频信号
传统的 LMS 滤波方法获得了更加优越的性能。通 的识别 [J]. 系统工程与电子技术, 2018, 40(7): 1449–1456.
Meng Qingsong, Wang Bin, Shao Gaoping. Recognition
过在实测数据上的测试，模型显示了在实测数据上 of underwater acoustic linear frequency modulation sig-
较好的泛化性，体现了一定的工程应用前景。 nals in α stable distribution noise[J]. System Engineering
and Electronics, 2018, 40(7): 1449–1456.
[9] 宋国丽, 郭新毅, 马力. 海洋环境噪声中的 α 稳定分布模
参 考 文 献 型 [J]. 声学学报, 2019, 44(2): 177–188.
Song Guoli, Guo Xinyi, Ma Li. α-stable distribution
model in ocean ambient noise[J]. Acta Acustica, 2019,
[1] 张安清, 邱天爽, 章新华. α 稳定分布的水声信号处理新方 44(2): 177–188.
法 [J]. 电子与信息学报, 2005, 12(8): 1201–1204. [10] 宋国丽. 南海北部海洋环境噪声时空差异与非高斯特性研
Zhang Anqing, Qiu Tianshuang, Zhang Xinhua. A new 究 [D]. 北京: 中国科学院大学, 2018.
underwater acoustic signal processing approach to α- [11] 张亚蕊. Alpha 稳定分布噪声衰落信道下调制信号识别 [D].
stable distribution[J]. Journal of Electronics & Informa- 西安: 西安电子科技大学, 2017.
tion Technology, 2005, 12(8): 1201–1204. [12] 吴国清. 周期性局部平稳过程双重谱图分析和测量 [J]. 声学
[2] Kounovsky T, Malek J. Single channel speech enhance- 学报, 1980, 12(2): 100–109.
ment using convolutional neural network[C]. Proceedings Wu Guoqing. Analysis and measurement of double spec-
of the 2017 IEEE International Workshop of Electronics, tra of periodic local stationary process[J]. Acta Acustica,
Control, Measurement, Signals and their Application to 1980, 12(2): 100–109.
Mechatronics (ECMSM), 2017. [13] 徐景峰, 韩树平, 舒象兰, 等. 非平稳噪声背景下瞬态声脉冲
[3] Park S R, Lee J. A fully convolutional neural network for 检测新方法 [J]. 声学学报, 2016, 41(3): 362–370.
speech enhancement[J]. Proceedings of the Annual Con- Xu Jingfeng, Han Shuping, Shu Xianglan, et al. A new
ference of the International Speech Communication Asso- method of transient acoustic pulse detection under the
ciation (INTERSPEECH), 2017: 1993–1997. nonstationary noise background[J]. Acta Acustica, 2016,
[4] Hao X, Shan C, Yong Xu, et al. An attention-based neu- 41(3): 362–370.
ral network approach for single channel speech enhance- [14] Glacier Bay National Park & Preserve. Sounds recorded
ment[C]// IEEE International Conference on Acoustics, in glacier bay[EB/OL]. [2019–02–19]. [2020–04–28].
Speech and Signal Processing (ICASSP), 2019. https://www.nps.gov/glba/learn/nature/soundclips.htm.