基于短时谱的噪声估计和语音增强研究.docVIP

基于短时谱的噪声估计和语音增强研究 摘要 噪声环境下，许多语音处理系统的性能急剧下降。语音增强作为解决噪声污染的一种有效的预处理技术，一直是语音信号处理领域中的热门课题。目的就是从带噪语音信号中尽可能地提取纯净的原始语音信号，提高信噪比改善语音质量 基于短时谱估计的语音增强方法简单适用信噪比范围大易于实时处理，是目前应用最广泛的方法。这种语音增强方法通常采用精确的噪声估计来采集噪声特性，通过良好的增强算法来得到期望的估计语音。本文立足于应用国内外的最新技术，深入系统地研究了短时谱增强方法中的噪声估计和语音增强这两大技术环节，主要实现的工作如下： （1）介绍语音增强和语音检测技术的研究背景、意义、存在的问题 （2）从语噪声估计方法基于统计模型VAD的方法与基于最小值约束的快速自适应方法。实验表明，VAD方法计算量小，易于实现，但是对非平稳噪声跟踪力度不够；基于最小值约束的方法能及时地跟踪噪声变化，获得准确的噪声估计，有效改善增强效果 （3）在短时谱估计的基础上，对谱减法进行了研究。 关键词：噪声估计语音检测Abstract Speech enhancement as a effective preprocessing technology to mitigate noise pollution, in which the speech processing systems sharply slowdown in performance, has been a hot topic. Its purpose is to remove all kinds of interference noise, enhance the SNR, and resume the original speech as purely as possible. Speech enhancement based on short-time spectral estimation is the most popular method which is simple, applicable in a wide range of SNR, and apt to real-time processing. This method obtains the noise property by using accurate noise estimation and gains the expectancy speech in the performance of favorable enhancement algorithm. According to the latest technology at home and abroad, this paper does systematic research on Noise Estimation and Speech Enhancement, which are two major technical issues of short-time spectral enhancement. The main work is listed as follows: 1. It introduces the research background, significance and existed problems of Voice Activity Detection and speech enhancement. 2. Starting with VAD and continually updated noise spectrum, it discusses the issue of noise estimation in speech enhancement system. Furthermore, it researches two effective methods to estimate noise. The one is based on statistical model VAD; the other one is based on the fast self-adaptive algorithm with constrained minimization. And the experiment shows that the former method whose computation is small is easy to realize, but cant follow up non-stationary noise timely, while the latter is very accurate in no