题名频域自适应噪声消除中的若干关键技术研究
作者覃波
学位类别博士
答辩日期2009-05-23
授予单位中国科学院声学研究所
授予地点声学研究所
关键词自适应噪声消除 频域分块自适应滤波器 步长控制 后滤波 DSP定点化 采样率差异
其他题名Study on frequency domain adaptive noise cancelation
学位专业信号与信息处理
中文摘要基于自适应噪声消除的语音增强算法,在免持电话、远程会议系统、蓝牙耳机式送话器、助听器等通信终端和电子设备中,有着广泛的应用。本文中,我们针对算法在实际应用中存在的问题,一方面深入研究了自适应滤波器的迭代更新,另一方面,我们致力于将自适应滤波和后处理单通道语音增强结合起来,从而构建一个完整的自适应噪声消除系统,并针对实际应用中可能出现的采样率不一致问题,提出了一个解决方案。 依托双通道自适应噪声消除这一应用背景,我们选择频域分块的滤波器结构,并着力构建一套鲁棒的频域自适应算法。首先,我们介绍了频域块更新LMS 算法(BLMS)和频域分块自适应算法(PBFDAF);然后,重点讨论了PBFDAF算法分频带更新的步长控制问题,推导出了一个理想的步长因子表达式,并给出了该步长因子的估计方法,此外,引入了双滤波器结构,辅助控制滤波器的迭代更新,以保证在cross-talk情况下,滤波器仍然保持较小的稳态误差和良好的鲁棒性;对自适应滤波系统的残余噪声,我们采用了一个基于音频掩蔽效应的后处理滤波算法对其进行抑制,通过保留一部分听觉阀值之下的噪声信息,保持了原始语音的质量,并避免了音乐噪声;然后,我们在TMS320C6416 DSP定点平台上实现了上述自适应噪声消除算法;最后,针对实际应用中可能出现的两通道信号间实际采样率不一致的情况,我们提出了一个基于相关性的采样率差异估计方法,在一个频率范围内对两相关信号进行采 样率差异的估计。
英文摘要Adaptive noise cancellation (ANC), which is an approach for speech enhancement, is widely used in teleconference systems、hands-free phones、Bluetooth headsets、hearing aids and other voice communication terminals. In this paper, we focus on finding a robust solution for these applications. On one hand, we discuss the problem of step-size control, which is the key component of ANC. On the other hand, we combine adaptive filtering technology and single channel speech enhancement, so as to construct a full ANC system. As well, we solve the problem of sample rate misalignment, which maybe a problem in some applications. In our application, the partitioned block frequency domain adaptive filter is selected, and we try to design a robust algorithm in frequency domain. Firstly, adaptive filtering in frequency domain and the structure of partitioned block filters are introduced; secondly, we drive an optimal bin-wise variable step-size and propose an estimation method, as well we bring a two filter structure to help control the updating of filter coefficients in presence of cross-talk. Thirdly, we take a psychoacoustic motivated weighting rule for post processing, which is supposed to reduce noises above auditory threshold and leave some noises masked by speech. This post processing method successfully avoids musical noise and retains good speech quality; Fourthly, we implement this algorithm on a fixed DSP platform of TMS320C6416. Finally, we design an approach which is based on coherence to solve the problem of sample rate misalignment between two related signals, with this approach, the sample rate difference can be estimated.
语种中文
公开日期2011-05-07
页码100
内容类型学位论文
源URL[http://159.226.59.140/handle/311008/558]  
专题声学研究所_声学所博硕士学位论文_1981-2009博硕士学位论文
推荐引用方式
GB/T 7714
覃波. 频域自适应噪声消除中的若干关键技术研究[D]. 声学研究所. 中国科学院声学研究所. 2009.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace