Energy-Aware Multichannel Noise Reduction in Wireless Acoustic Sensor Networks (面向无线声学传感器网络的多通道语音增强)

发布者:计算机信息发布时间:2019-04-02浏览次数:13

报告题目:Energy-Aware Multichannel Noise Reduction in Wireless Acoustic Sensor Networks (面向无线声学传感器网络的多通道语音增强)

报告人:张结 博士

单位:北京大学

报告时间:201943日(周三)下午14:00

报告地点:翡翠科教楼A1106


报告人简介:张结,受国家留学基金委资助公派于荷兰代尔夫特理工大学攻读博士学位。研究方向主要是:语音增强、声源定位、双耳声学、传感器网络、凸优化等。担任IEEE/ACM TASLPIEEE Trans. SignalProcessing, IEEE Communication/Signal Processing LettersICASSPICRAIROS 等国际刊物的审稿人。目前已发表学术论文十余篇,其中被SCI收录8篇,主要发表于IEEE/ACM TASLPIEEE Trans. SignalProcessing,并获得2018 10th IEEE SensorArray and Multichannel Signal Processing Workshop (SAM)最佳论文奖,申请并授权国家发明专利2项。

Jie Zhang is pursuing his PhD at Delft University of Technology (TU Delft) in Netherlands,funded by the China Scholarship Council. His research interests cover multichannel speech enhancement, source localization, binaural auditory, wireless sensor networks, convex

optimization, etc. He serves as a reviewer for many international conferences/journals, e.g.,IEEE/ACM TASLP, IEEE Signal Processing Letters, IEEE Communication Letters, ICASSP,ICRA,IROS, etc. He has published more than 10 academic articles, and 8 are indexed by SCI. He received the Best Student Paper Award from 2018 10th IEEE Sensor Array and Multichannel Signal Processing Workshop (SAM). He holds two national patents.


摘要:近些年,无限声学传感器设备的使用越来越普遍,我们无时不处在一个大规模传感器网络里面。相对于传统的麦克风阵列,无线声学传感器网络具有很多优势,比如不受限于应用平台的体积、易于布局、空间感知范围更广等。同时,也给语音信号处理带来了新的挑战,比如网络资源(电池能量、通信带宽)消耗、时变的拓扑结构、设备时间同步等。对此,我们首先提出了如何选择能量消耗最低的麦克风子集来达到指定的语音降噪效果;其次,由于每个设备在传输数据之前要经过量化,我们提出了如何最优地分配网络量化速率;再次,考虑到中心化网络鲁棒性差的问题,我们提出了分布式语音降噪算法;最后,由于传统基于波束形成的语音降噪算法需要用到声学传递函数,我们提出了一种低通信速率场景下的声学传递函数估计方法。

Abstract: Nowadays, wireless acoustic devices are more and more often-used in our daily life, e.g., smartphones, laptops, hand-free telephony kits, hearing aids, resulting in a large-scale wireless acoustic sensor network (WASN). Compared to the conventional microphone arrays, the so-called WASN has many advantages, e.g., the devices can be placed at any locations where it might be difficult to position wired microphones, a larger acoustic scene can be sampled. Meanwhile, there are also some new challenges that need to be addressed in the context of WASNs, e.g., resource consumption, time-varying network topology, synchronization, etc. For these, we first proposed a sensor selection strategy, which is obtained by minimizing the total transmission power over the WASN subject to a constraint on an expected noise reduction performance. Second, we proposed to distribute the communication rates to the sensors, which are used for quantizing the raw sensor measurements. Third, to improve the robustness against the network variation, we proposed a distributed noise reduction algorithm. Finally, since the traditional beamforming based noise reduction algorithms are based on the acoustic transfer function (ATF), we proposed a method for estimating this ATF at low communication rates.