基于显式增益阈值的参数维纳滤波人工耳蜗降噪算法

发布时间：2018-02-01 19:33

本文关键词： 人工耳蜗降噪算法增益函数显式增益阈值　出处：《深圳大学》2017年硕士论文　论文类型：学位论文

【摘要】：人工耳蜗(Cochlear implant,CI)是目前能使重度感音性听力障碍患者恢复听觉感知的唯一装置。它通过言语处理器对声音信号提取包络信息,经非线性压缩和振幅调制等处理,将声音信号转换为相应电极的电信号刺激残留的听觉神经,使患者恢复听力。它在安静环境下能为CI植入者提供较好的言语感知体验,然而在噪声环境下,大多数CI产品无法为植入者提供足够的频率分辨率、时域精细结构等信息,使CI植入者的言语识别率出现明显的下降,严重影响他们在噪声环境下的正常交流。因此,针对CI植入者在噪声下言语感知差的问题开展降噪算法的研究具有重要意义。针对该问题,本文提出一种基于显式增益阈值(apparent gain threshold,aGT)的参数维纳滤波降噪算法。本文首先对人工耳蜗的背景知识、国内外人工耳蜗降噪算法的发展和研究现状进行叙述;其次重点介绍了Mauger提出的基于信噪比和参数维纳增益函数的降噪算法,并针对该算法存在主要问题--需要经过大量的测试实验才能确定增益函数最优参数---提出一种改进算法。通过实验验证分析aGT曲线与最优参数(、)的关系,提出基于aGT的参数维纳滤波降噪算法,将二维平面的最优参数(、)寻找问题简化为单个参数aGT的最优化问题,从而极大地减少了测试者的参数调校实验量。最后,本文通过声码器仿真声实验和CI植入者实验验证所提算法的降噪性能。由仿真声实验结果可知,大多数测试者在aGT为5dB时能获得较好的言语识别率,不同通道数的选择对于降噪算法的性能影响不大。与其他三种降噪方法的识别率比较发现,当增益阈值为5dB时,本文提出的算法能达到与方法I和II相当的降噪效果,而方法I和II中噪声估计是在噪声信号已知的情况下得到的。对比结果也表明基于感知特性的参数维纳增益函数比二值掩蔽增益函数更适合CI植入者。从CI植入者的实验结果可知,与正常听力者相比,CI植入者更偏向于选择大于0dB的增益阈值。同样地,在aGT值为5dB时,大多数测试者可以得到较高的单词识别率。
[Abstract]:Cochlear implant. At present, CI is the only device that can restore auditory perception in patients with severe sensorineural hearing impairment. It extracts envelope information from sound signal by speech processor and is processed by nonlinear compression and amplitude modulation. The sound signal can be converted into the corresponding electrode electrical signal to stimulate the residual auditory nerve, so that the patient can recover hearing. It can provide a better speech perception experience for CI implants in the quiet environment, but in the noise environment. Most CI products can not provide enough frequency resolution, time domain fine structure and other information for implants, so the speech recognition rate of CI implants decreased significantly. Therefore, it is of great significance to study the speech perception of CI implants under noise. In this paper, a parameter Wiener filter denoising algorithm based on explicit gain threshold gain threshold is proposed. Firstly, the background knowledge of cochlear implant is given. The development and research status of cochlear denoising algorithms at home and abroad are described. Secondly, the noise reduction algorithm based on signal-to-noise ratio and parameter Wiener gain function proposed by Mauger is introduced. Aiming at the main problem of the algorithm, which needs a lot of tests to determine the optimal parameter of the gain function, an improved algorithm is proposed. The aGT curve and the optimal parameter are analyzed by the experiment. A parameter Wiener filter denoising algorithm based on aGT is proposed, which simplifies the problem of finding the optimal parameters of two-dimensional plane to the optimization problem of single parameter aGT. In order to greatly reduce the parameters of the tester calibration experiment. Finally, the noise reduction performance of the proposed algorithm is verified by the vocoder simulation sound experiment and CI implant experiment. The result of the simulation acoustic experiment shows that the proposed algorithm can reduce the noise of the proposed algorithm. Most of the testers can obtain better speech recognition rate when aGT is 5 dB, and the choice of different channel numbers has little effect on the performance of the noise reduction algorithm. Compared with the other three noise reduction methods, it is found that the recognition rate of the proposed algorithm is better than that of the other three methods. When the gain threshold is 5 dB, the proposed algorithm can achieve the same denoising effect as that of methods I and II. In methods I and II, the noise estimation is obtained when the noise signal is known. The comparative results also show that the parameter Wiener gain function based on perceptual characteristics is more suitable for CI implants than the binary masking gain function. The experimental results of I implants showed that. Compared with the normal hearing group, CI implants were more inclined to choose a gain threshold greater than 0 dB. Similarly, when the aGT value was 5 dB, most of the subjects could get a higher word recognition rate.
【学位授予单位】：深圳大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN912.3;TN713

【参考文献】