（毕业学术论文设计）-噪声环境下语音端点检测技术的开发.docVIP

噪声环境下语音端点检测技术的开发 上海交通大学网络学院 PAGE PAGE 1 噪声环境下语音端点检测技术的开发 摘 要 语音识别技术就是让机器通过识别和理解过程把语音信号转变为相应的文本或命令的高技术。在实际应用中，语音识别系统的性能会受到各种因素的影响。影响语音识别能力的一个重要因素是语音端点检测的准确性。在实验室条件下普通的语音识别系统能达到较高的识别率，但如果有一定的背景噪声，特别是强背景噪声存在的应用场合，性能将会急剧下降。因此，如何提高语音识别系统在噪声环境下的鲁棒性，是语音识别研究中的一个重要的课题。 本文就噪声环境下语音端点检测技术的开发进行了探讨，提出了一种基于信号的相关函数的语音端点检测法，它采用了极性相关法进行计算并提供了相应的关键算法，通过对语音识别实验的比较，考察了该算法的有效性和实用性。与传统的语音端点检测法相比，本文提出的端点检测法能使语音/非语音成份估计及分离变得更为有效和方便，同时由于极性相关法采用取小累加处理法从而使得端点检测运算量可大为减少。 本文由四部分组成： 第一章引言。阐述了课题的研究背景与意义， 第二章语音信号处理和端点检测。介绍了语音信号处理和端点检测中的一些基本概念、原理和方法。 第三章基于信号的极性相关函数的语音端点检测法。这是本文的核心部分，着重对基于极性相关法的带噪语音端点检测进行了讨论，具体探讨了极性相关法的关键算法，给出部分统计数据和结果总结。 第四章总结和展望。 [关键词]:语音/非语音检测,极性相关,类似度,语音识别 Realization of Speech/Non-Speech Detection in Adverse Conditions Abstract： Speech/non-speech detection is important in many areas of speech processing technology. In real environments, speech signal is usually subject to environmental factors such as background noise、speaker individuality and voice distortion, etc, which sharply hurt their performance. The speech/non-speech detection algorithm greatly affect the performance of speech processing systems. Therefore, the task of improving the robustness of speech recognition system in real environment makes great sense, being an important task of automatic speech recognition research. In this paper a high performance mandarin Speech/non-speech detection system is given. The algorithm is based on polarity correlation method. We use a similar degree between the positive polarity correlation sequence and the negative polarity correlation sequence for the input speech signal to decide on the speech/non-speech section. A practical algorithm is proposed, which can detect for Boundary Detection. The evaluation results show this algorithm can obtain higher rates of accuracy in noisy environments. Having adopted the above-mentioned algorithm, the computation of Speech/non-speech detection s shift matching can be greatly compressed., This paper is organized