【作者】 李晋；

【导师】 王玲；

【作者基本信息】 湖南师范大学 ， 电路与系统， 2006， 硕士

【摘要】 语音信号处理中,端点检测是指从背景噪声中准确地检测出语音信号的起始点和终止点,从而在语音识别中提高识别精度以及减少识别时间,在语音编码中降低噪声的比特率,提高编码效率。由于现实环境下各种不确定噪声的引入通常会使检测性能显著下降,因此,低信噪比下语音端点检测技术的研究意义非常重要。 本文首先总结了现有典型的语音端点检测算法,分析了其中几种端点检测算法所选用的特征,给出了仿真结果和一些改进。随后提出了噪声环境下两种语音端点检测新算法。算法一:从基于人耳的听觉系统出发,对Mel标度滤波器组进行研究,提出了语音信号的一种新的自适应时频参数,该参数既考虑了声道响应,又符合人耳听觉特性,仿真结果表明了它的优越性。算法二:结合抗噪性能好的Mel倒谱距离和多带能量熵特征提出了一种改进的孤立词端点检测算法,该算法不需要估计背景噪声来调整门限阈值,仿真结果表明在常见的噪声环境下效果较好,算法实现简单,环境适应性较强。文章最后对本文所提的两种算法进行了总结,提出了一些在今后工作中需要改进的问题,并对近几年来出现的一些研究新动向作了简单的介绍和展望,指出了端点检测未来的发展前景。更多还原

【Abstract】 Speech endpoint detection, which accurately detects the beginning and ending of the speech signal from background noise, is crucial for the speech signal processing. The accurate endpoint detection can improve the speech recognition accuracy with less recognition time, and decrease the noisy bit rate. Moreover, the efficiency of the speech coding can be increased by endpoint detection. Generally speaking, the performance of the detection severely degrades in the actual noisy condition. Therefore, the speech endpoint detection is particularly important in the speech recognition, especially for low SNR. Firstly, with the analysis of some speech features, the existing typical algorithms of endpoint detection are summarized. And some simulation results and improvements are also shown in the paper. Then, two new algorithms of the speech endpoint detection are proposed. The first one: based on the hearing system of the human, a new adaptive time frequency parameter considers both auditory and vocal tract character is proposed by studying the Mel filter bank. Simulation results have shown the performance of the algorithm is better than other existing algorithms. The second one: combined with the Mel cepstrum distance and the multi-band energy entropy, animproved isolated word endpoint detection algorithm is proposed, which can exempt the adjustment of the decision threshold by means of estimating background noise. Simulation results show the algorithm has better robust capability , and high efficiency can be obtained in common noisy environments. At the end of the article, the two proposed algorithms are summarized. Some new study fields within the past two years are introduced and developing perspective of endpoint detection is referred to.更多还原