语音端点检测算法的研究及matla程序仿真与实现（本科）毕业论文.docVIP

摘要 语音端点检测是指从一段语音信号中准确的找出语音信号的起始点和结束点，它的目的是为了使有效的语音信号和无用的噪声信号得以分离，因此在语音识别、语音增强、语音编码、回声抵消等系统中得到广泛应用。 目前端点检测方法大体上可以分成两类，一类是基于阈值的方法，该方法根据语音信号和噪声信号的不同特征，提取每一段语音信号的特征，然后把这些特征值与设定的阈值进行比较，从而达到语音端点检测的目的，该方法原理简单，运算方便，所以被人们广泛使用，本文主要对基于阀值的方法进行研究。另一类方法是基于模式识别的方法，需要估计语音信号和噪声信号的模型参数来进行检测。由于基于模式识别的方法自身复杂度高，运算量大，因此很难被人们应用到实时语音信号系统中去。 端点检测在语音信号处理中占有十分重要的地位，直接影响着系统的性能。本文首先对语音信号进行简单的时域分析，其次利用短时能量和过零率算法、倒谱算法、谱熵算法进行语音端点检测，并对这几种算法进行端点检测，并进行实验分析。本文首先分别用各算法对原始语音信号进行端点检测，并对各算法检测结果进行分析和比较。其次再对语音信号加噪，对不同信噪比值进行端点检测，分析比较各算法在不同信噪比下的端点检测结果，实验结果表明谱熵算法语音端点检测结果比其他两种方法好。 关键词　语音端点检测；语音信号处理；短时能量和过零率；倒谱；谱熵 Abstract Endpoint detection is a voice signal from the accurate speech signal to the identify start and the end points, the purpose is to enable to separated the effective voice signals and un-useful noise. So, in the speech recognition system, speech enhancement, speech coding, echo cancellation and other systems are widely used. In Current the endpoint detection can be roughly divided into two categories, one is based on the threshold method based on the different characteristics of speech signal and the noise signals, a voice signal for each extracted feature, and then set the values ??of these thresholds compare with the values ??to achieve the endpoint detection purposes, This method is simple, it convenient operation, it is widely used, the main in this paper is based on the method of threshold method. Another method is based on the method of pattern recognition , it needs to estimate the speech signal and the noise signal model parameters were detected. Because is based on the method of pattern recognition and high self-complexity, a large amount of computation, so it is difficult to be use in real-time voice signal system for people. The Endpoint detection is take a very important position in the speech recognition, it directly affects the performance of the system. In this article first domain analysis in simple speech signal time, than dual threshold algorithm, cepstrum algorithm, spectral