论文标题:语音增强技术的研究及应用
Research and Application for Technology of Speech Enhancement
论文作者 姜琳峰
论文导师 孙洪,论文学位 硕士,论文专业 通信与信息系统
论文单位 武汉大学,点击次数 85,论文页数 65页File Size10315k
2004-04-01论文网 http://www.lw23.com/lunwen_174216892/ 语音增强;噪声估计;TURBO迭代滤波器
Speech Enhancement;Noise Estimate;TURBO Iteration Filter
由于人们在语音通信过程中不可避免地会受到来自周围环境、传输媒介引入的噪声，通信设备内部电噪声，乃至其它讲话者的干扰。这些干扰最终将使接收者接受到的语声已非纯净的原始语音信号，而是受噪声污染的带噪语音信号。例如，安装在汽车、飞机或舰船上的电话，街道、机场的公用电话以及现在人们常用的移动电话，常受到很强背景噪声的干扰，严重地影响通话质量。因此，语音增强技术就成为数字语音信号处理中的重要组成部分。 所谓语音增强就是对带噪语音进行处理，以改善语音质量。提高语音的清晰度、可懂度和舒适度，使人易于接受或提高语音处理系统的性能。大多数实用的语音增强系统均是由精确的噪声估计器和良好的滤波器来实现对带噪语音进行去噪。 传统的噪声估计方法是基于语音活性检测(Voice Activity Detection，VAD)的，具有简单，易实现的优点。但是实验证明，VAD在低信噪比的情况下对噪声特性的精确估计有很大的缺陷。而基于连续更新噪声谱的噪声估计方法，既可以省略VAD，又能较大地提高带噪语音的噪声估计精度，尤其在非平稳噪声或者是波动噪声(Fluctuating Noise)的噪声特性及时更新方面有良好的表现，在实际应用中取得了很好的效果。 在滤波器的设计方面，传统的基于短时谱分析的方法具有易实现，实用性强的特点。在数十年的研究中，新兴的方法层出不穷，有的借助于人耳的听觉特性，有的借助于新兴的数学工具，例如小波分解、隐马尔可夫模型(HMM)等等。但是很多方法离实用阶段还有很长的距离。同时，基于TURBO迭代的滤波思想有了很大的发展，尽管TURBO迭代为什么具有这么良好的滤波效果目前还是没有人能解释，但是这种迭代思想在编译码、图像滤波等各个领域都得到了很好的应用。于是本论文将TURBO迭代的滤波思想引入到语音增强算法中，提出TURBO语音增强算法，并得到了令人满意的结果。 本论文研究的是语音增强技术及其应用，并提出TURBO语音增强算法。下面是本文的主要工作： 1．对基于VAD的噪声估计方法和连续更新噪声谱的噪声估计方法进行了研究及比较，针对仿真结果分析了两类噪声估计方法的性能，并将两者结合起来，做出改进，应用于实际的增强系统。 2．系统地研究了各种语音增强方法，包括基于短时谱分析的语音增强算法、基于语音生成模型的语音增强算法、基于人耳听觉掩蔽效应的语音增强方法等。在此基础之上，提出了一种基于TURBO迭代滤波器的TURBO语音增强方法，并对各种算法进行了分析和比较。 3．在对噪声估计算法和语音增强算法的研究基础上，结合连续更新噪声谱的噪声估计方法和基于短时谱分析的语音增强算法设计了两种实用语音增强系统，针对仿真语音和真实语音均做出结果和分析，并应用于实际项目中。
In the process of speech communication, the speech signal is often corrupted acoustically by ambient noise, electronic noise, or speech of other talkers. So a received speech signal is not pure, but noisy speech signal. For example, a speech signal caused by phones installed on plane, car or ship, public phones in the street or mobile phone is often disturbed by strong background noise. Therefore, speech enhancement technique plays a great pan of digtal speech signal processing.Speech enhancement is a process of filtering the noisy speech signal to improve its quality and performance of speech processing system, and to make speech clearer, more understood and more comfortable. Most practical speech enhancement systems composed of an accuracy noise estimator and good filter to reduce the noise.Classical methods of estimating noise are based on Voice Activity Detection (VAD) which is easy to be realized in practical. But the experiment results show that VAD has poor perfermance in the side of accurate estimate of noise. Another methods based on continuous renewal noise spectral greatly improve the accuracy of noise estimator without VAD. Particlularly, the latters have the excellent performance for fluctuating noise and obtain the good results in practical application.In the design of filter, classical methods are based on Short Time Spectral Analysis (STSA). In the past decades, many new methods are proposed, such as speech enhancement based on wavelet or Hidden Markov Model (HMM), acoustical characteristic, and so on. While these methods have still a long way for practical application. Recently, the principle of TURBO iteration idea developped rapidly in many fields: encoding, decoding, image filter, etc. In the thesis TURBO speech enhancement method based on TURBO iterative algorithm is proposed. The result obtained by computer simulation is satisfactory.The thesis places emphesis on research and application of speech enhancement technology, and proposes TURBO speech enhancement method, Following is the main work of this thesis:1. A variety of methods of noise estimate are discussed. And a comparation and the evalution of the results are given. An improved method based on the connection of two kinds of noise estimate is proposed.2. A variety of methods about speech enhancement are discussed. And TURBO speech enhancement based on TURBO iteration is propsed.3. Two speech enhancement systems composed of noise estimator and filter design based on STSA are designed, and the results for both simulated and real speech signal are given in the thesis.
|
