Environmental noise suppression algorithms for hearing aids

助听器环境噪声抑制算法

基本信息

批准号：
07457395
负责人：
SONE Toshio
金额：
$ 3.65万
依托单位：
TOHOKU UNIVERSITY
依托单位国家：
日本
项目类别：
Grant-in-Aid for Scientific Research (B)
财政年份：
1995
资助国家：
日本
起止时间：
1995 至 1996
项目状态：
已结题

项目摘要

Digital signal processing algorithms to suppress environmental noise were studied for advanced hearing aid. Algorithms with one microphone and two microphones were examined.As to digital signal processing algorithm to suppress environmental noise with one microphone, the spectral subtraction method is well known and widely used. In the method, averaged frequency spectrum of environmental noise is subtracted from the frequency spectrum of input signal. This method usually uses Fourier transform to calculate the frequency spectrum. The conventional spectrum subtraction with Fourier transform, however, gives insufficient performance. We, therefore, used wavelet transform, which was considered to be similar to the frequency analysis process in the auditory system. Using various kinds of basis functions for the wavelet analysis, hearing experiments were carried out to evaluate their performances. The result showed that the intelligibility was better restored with non-orthogonal basis functions than with orthogonal ones.Digital signal algorithms with two microphones were examined next. When multiple microphones are used, beam-forming techniques are the most promising. Though conventional beam-forming technique is effective for a free or almost free sound field, it is not very effective in a live or reverberant room. Since such deterioration is caused by the phase relation between of direct sound and reflections, we developed a new beam-forming algorithm which did not use phase relation of the input signal but focused only on its power spectrum. The new algorithm showed better performance in a dispersive sound field than the conventional alrorithm.

研究了抑制环境噪声的数字信号处理算法以进行高级助听器。检查了带有一个麦克风和两个麦克风的算法。由于数字信号处理算法以用一个麦克风抑制环境噪声，光谱减法方法是众所周知的，并且广泛使用。在该方法中，从输入信号的频谱中减去环境噪声的平均频率频谱。该方法通常使用傅立叶变换来计算频谱。然而，传统的光谱减法具有傅立叶变换的性能不足。因此，我们使用的小波变换被认为与听觉系统中的频率分析过程相似。使用各种基础函数进行小波分析，进行了听力实验以评估其性能。结果表明，与正交函数相比，通过非正交函数更好地恢复了可理解性。接下来检查了带有两个麦克风的数字信号算法。当使用多个麦克风时，最有希望的技术是最有前途的技术。尽管传统的梁形成技术对于免费或几乎免费的声场有效，但在现场或混响的房间中并不是很有效。由于这种恶化是由直接声音和反射之间的相位关系引起的，因此我们开发了一种新的梁形成算法，该算法不使用输入信号的相位关系，而仅专注于其功率谱。与传统alrithm相比，新算法在色散声场中显示出更好的性能。