数字助听器中语音增强方法的研究

发布时间：2018-05-25 06:37

本文选题：数字助听器 + 语音增强　；参考：《东南大学》2016年硕士论文

【摘要】：受人口结构老龄化、环境噪声污染等因素的影响,越来越多的人听觉系统受到了损伤。在听力损失患者的听力矫正治疗中,佩戴助听器是最安全、最有效的方式,但噪声污染问题严重影响了助听器的实际使用效果,且较之听力正常人而言,听损患者在噪声环境中的语音理解能力本身就更低。语音增强技术能有效消除背景噪声,改善语音质量,从而提高患者在噪声环境中的语音理解度。在国家自然科学基金(61301219)和江苏省自然科学基金(BK20130241)的资助下,本文在深入理解和研究已有算法的基础上,重点研究了适用于数字助听器的语音增强技术,主要研究工作包括以下几个部分：(1)介绍了维纳滤波语音增强技术,重点研究了基于先验信噪比(Priori SNR)估计的维纳滤波法。针对基于Priori SNR估计的维纳滤波未能有效提高语音可懂度的缺点进行了研究与改进：首先,引入了以MMSE为准则的两步先验信噪比估计取代原“直接判决”法；其次,针对信噪比为负的区域,增益函数的过高估计会严重降低语音可懂度的问题对增益函数进行了改进,放大了负信噪比区域的噪声谱,从而降低增益函数被过高估计的可能性。与原方法的对比实验验证了改进的方法能够有效的提高语音可懂度。(2)针对基于Priori SNR估计的维纳滤波算法计算复杂度较高的缺点,提出了一种计算复杂度低的数字助听器子带语音增强算法,详细介绍了该算法的原理及具体实现。该方法利用计算各子带信号功率取代传统方法中的功率谱的计算,省去了信号时频域的转换,极大程度的降低了算法的计算复杂度,满足了数字助听器对实时性和低功耗的要求。并通过仿真实验将该算法与改进谱减法和基于Priori SNR估计的维纳滤波法进行了对比分析,验证了算法的有效性。(3)研究了基于方向性麦克风的语音增强技术,介绍了方向性麦克风用于语音增强的原理及实现方式,重点对一阶自适应方向性麦克风的原理结构进行了研究。方向性麦克风只能进行初步的噪声过滤,而子带语音增强算法对信噪比过低的信号进行增强处理时会造成较大的语音失真,针对上述问题,提出了一种方向性麦克风与子带语音增强相结合的两步语音增强算法,首先由方向性麦克风初步提升信噪比,再由子带语音增强方法作进一步处理,两者的结合有效的弥补了彼此的不足,使得最终增强的语音信号质量更好。仿真实验的结果验证了提出算法的有效性。
[Abstract]:Due to the aging of population structure and environmental noise pollution, more and more people's hearing system is damaged. It is the safest and most effective way to wear hearing aid in hearing loss patients, but the problem of noise pollution seriously affects the actual effect of hearing aid. The speech comprehension ability of hearing impaired patients in noise environment is even lower. Speech enhancement technology can effectively eliminate background noise, improve speech quality and improve the speech comprehension of patients in noisy environment. Supported by the National Natural Science Foundation of China 61301219) and the Natural Science Foundation of Jiangsu Province BK20130241), based on the deep understanding and research of the existing algorithms, this paper focuses on the speech enhancement technology suitable for digital hearing aids. The main research work includes the following parts: 1) the Wiener filter speech enhancement technique is introduced, and the Wiener filtering method based on Priori SNR estimation is emphatically studied. The shortcomings of Wiener filter based on Priori SNR estimation to improve speech intelligibility are studied and improved. Firstly, a two-step prior SNR estimation based on MMSE criterion is introduced to replace the original "direct decision" method. For the region with negative SNR, the gain function can be greatly reduced by overestimation of the gain function, and the noise spectrum of the negative SNR region is enlarged, thus reducing the possibility that the gain function can be overestimated. The comparison experiment with the original method proves that the improved method can effectively improve the speech intelligibility. (2) aiming at the disadvantage of high computational complexity of Wiener filter algorithm based on Priori SNR estimation, the improved method can improve speech intelligibility effectively. A speech enhancement algorithm for digital hearing aids with low computational complexity is proposed. The principle and implementation of the algorithm are introduced in detail. In this method, the power spectrum of each subband signal is calculated instead of the power spectrum of the traditional method, and the time-frequency domain conversion of the signal is eliminated, and the computational complexity of the algorithm is greatly reduced. It meets the requirement of real-time and low power consumption for digital hearing aid. The algorithm is compared with the improved spectral subtraction method and the Wiener filter method based on Priori SNR estimation, and the effectiveness of the algorithm is verified. The speech enhancement technology based on directional microphone is studied. The principle and implementation of directional microphone for speech enhancement are introduced, and the principle structure of the first order adaptive directional microphone is studied. Directional microphones can only carry out initial noise filtering, but subband speech enhancement algorithm will cause large speech distortion when processing signals with too low SNR. In view of the above problems, A two-step speech enhancement algorithm combining directional microphone and subband speech enhancement is proposed. Firstly, the signal to noise ratio (SNR) is initially enhanced by directional microphone, and then further processed by subband speech enhancement method. The combination of the two effectively makes up for each other's shortcomings, making the final enhanced speech signal quality better. Simulation results show the effectiveness of the proposed algorithm.
【学位授予单位】：东南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.35;TH785.1

【参考文献】