计算机工程 ›› 2011, Vol. 37 ›› Issue (22): 5-7.doi: 10.3969/j.issn.1000-3428.2011.22.002

• 专栏 • 上一篇    下一篇

基于循环自相关函数的浊音端点检测

李 皓 1,2,唐朝京 1   

  1. (1. 国防科学技术大学电子科学与工程学院,长沙 410073;2. 中国人民解放军75753部队,广州 510600)
  • 收稿日期:2011-07-07 出版日期:2011-11-18 发布日期:2011-11-20
  • 作者简介:李 皓(1982-),男,博士研究生,主研方向:多媒体通信,可视语音合成;唐朝京,教授、博士生导师
  • 基金项目:

    国家部委基金资助项目

Voiced Sound Endpoint Detection Based on Circular Autocorrelation Function

LI Hao 1,2, TANG Chao-jing 1   

  1. (1. College of Electronic Science and Engineering, National University of Defense Technology, Changsha 410073, China; 2. Troop 75753, The Chinese People’s Liberation Army, Guangzhou 510600, China)
  • Received:2011-07-07 Online:2011-11-18 Published:2011-11-20

摘要:

为提高浊音端点检测的准确率和效率,提出一种基于循环自相关函数的检测方法。设计语音的循环自相关函数,利用该函数与短时能量定义状态及转移损失函数,通过动态规划方法判别浊音的端点,并采用不同分类判断方法与检测函数进行测试。实验结果表明,与基于能量及谱墒的方法相比,该方法的抗噪性能较好。

关键词: 浊音, 端点检测, 循环自相关函数, 短时能量, 动态规划, 损失函数

Abstract:

To enhance the accuracy and efficiency of endpoint detection, a detection method based on Circular Autocorrelation Function(CACF) is proposed. The method calculates CACF of the speech, defines the loss functions of state and state transforming with the values of CAF and the short-term energy, and decides the voiced endpoints with dynamic programming. Experimental results display several comparisons including that among detections using the traditional Autocorrelation Function(CAF), average magnitude difference function and the CACF, which demonstrate that CACF improves the accuracy and efficiency of endpoint detection and better resists the acoustic noise than the traditional energy and spectral entropy method.

Key words: voiced sound endpoint detection, Circular Autocorrelation Function(CACF), short-term energy, dynamic programming, loss function

中图分类号: