A Study on Noise Suppression using Perceptual Weighting Approach based on Auditory Masking Effect.

为了提高受加性背景噪声污染的语音质量，本研究的目的是通过基于听觉掩蔽效应的感知加权函数的谱减法方法来获得去除噪声后的感知良好的输出语音，我认为这项研究是沿着开始的计划进行的。在第一年中，通过将感知加权函数引入噪声抑制系统并评估其性能，获得了结果-1。在第二年，通过自适应地控制比例因子获得了结果-2。此外，本研究的结果在国际会议和国内会议或学会上发表，并发表在IEICE杂志上。提出了一种基于感知频率加权的噪声频谱抑制算法，该算法通过对输入语音的频谱包络进行感知，得到感知尺度函数，从而达到抑制背景噪声的目的.（1）提出了一种新的谱包络估计方法，该方法通过一次简单的谱减法从信号中计算线性预测系数，提高了谱包络估计的精度。输出信号的质量得到了降低，噪声得到了改善. (2)该方法通过改变加权因子的值，从固定值（在第一年）到可变值（在下一年），自适应地控制加权因子的值。结果表明，该方法能准确地估计噪声谱。

In order to enhance a quality of speech corrupted by additive background noise, this research aims that perceptually good output speech removed the noise is achieved by the spectral subtraction approach based on a perceptual weighting function concerned with an auditory masking effect.I think that this research was approximately done along to the beginning plan. In the first year the result-1 was obtained by introducing the perceptual weighting function to the noise suppression system and by evaluating its performance. In the next year the results-2 were obtained by controlling the scaling factors adaptively. In addition, the results of this research were presented at the international conference and at the domestic meeting or society and contributed to the journal of IEICE.RESEARCH RESULTS1. Introducing of algorithm using perceptual frequency weighting to reduce the noise spectrumIt was proposed that the approach reduces the background noise by using the perceptual scaling function obtained from the spectral envelope of the input speech.2. Improvement of algorithm obtained from the first year's research(1) It was proposed that the approach improves the accuracy of the spectral envelope estimation by the fact that linear prediction coefficients are calculated from the signal which applies one time simple spectral subtraction method. The quality of the output signal reduced the noise was improved.(2) It was proposed that the approach controls adaptively the values of weighting factors by changing the weighting factors from the fixed value (in first year) to the variable value (in the next year). As a result of this approach the accurate estimation of noise spectrum was obtained.