文章摘要
安茂波,国雁萌,付强,颜永红.行驶汽车环境中的话音活动检测研究[J].,2012,31(5):387-392
行驶汽车环境中的话音活动检测研究
Voice activity detection for an in-vehicle environment
投稿时间:2012-01-19  
中文摘要:
      话音活动检测是语音交互和通信系统的重要部分,其作用是区分输入信号中的语音段和背景噪声段,检测的依据主要是语音和噪声的各种时频特性,其中,浊语音的周期性和谐波特性是一种广泛应用的特征。但是在行驶的汽车环境中,由于噪声非平稳且信噪比较低,这类特征较难得到可靠的检测。为此,本文根据浊音谐波结构的基本规律,利用时变噪声环境中各频带信噪比不同的特点,提出一种较为鲁棒的谐波快速检测算法。算法以较小的时频块为分析单元,利用一组基频在对数尺度上变化的谐波模板,自适应地搜索谐波结构清晰的部分,并以此检测浊语音信号。实验证明,该算法能够在行驶的汽车环境中达到较可靠的话音/非话音区别效果。
英文摘要:
      Voice activity detection (VAD) is very important for speech interaction and communication systems to discriminate the segments of speech and background noise. The speech/ non-speech classification is usually based on the time-frequency features of speech and noise, among which the periodicity and harmonic feature is widely used. However, these kinds of feature cannot be extracted reliably for the in-vehicle speech applications, where the signal-to-noise ratio (SNR) is low and the noise is non-stationary. A fast voice detection algorithm is proposed in this paper based on the harmonic-structure of voice. Several harmonic templates are generated with fundamental frequency varies in log-scale, and they are used to detect the clear harmonic structure in several small time-frequency blocks separately. The proposed voice activity detection algorithm shows good performance on the real in-vehicle database.
DOI:10.11684/j.issn.1000-310X.2012.05.011
中文关键词: 话音活动检测,谐波结构,信噪比,时变噪声
英文关键词: Voice activity detection, Harmonic structure, Signal-to-noise ratio, Time-varying noise
基金项目:国家自然科学基金(批准号: 10925419, 90920302, 61072124, 11074275, 11161140319); 中国科学院战略性先导科技专项(面向感知中国的新一代信息技术研究, 编号: XDA06030100)
作者单位
安茂波 国家计算机网络应急技术处理协调中心 北京 100029
 
国雁萌 中国科学院声学研究所 北京 100190 
付强 中国科学院声学研究所 北京 100190 
颜永红 中国科学院声学研究所 北京 100190 
摘要点击次数: 3109
全文下载次数: 724
查看全文   查看/发表评论  下载PDF阅读器
关闭