摘要
为在有色非高斯噪声背景下实时、有效地区分语音信号与各种背景噪声,提出了一种基于灰关联分析的语音激活检测(VAD)算法。该算法提取语音信号过零率、线性预测系数、倒谱系数和转移倒谱系数4种特征参数作为关联参数,通过跟踪语音与噪声灰关联度的变化确定判决门限,实现语音激活检测。仿真结果表明:该算法在无噪声背景中识别率为100%,在-5 dB噪声背景环境,识别率可达80%以上。此算法对有色非高斯背景噪声不敏感,而且计算简单、可靠性高,在语音激活检测中具有可行性。
To real-time classify voice and noise under non-stationary nongaussian noise effectively, a voice activity detection (VAD) method based on the gray correlation analysis algorithm was proposed by analyzing ZCR(zero crossing rate), LPC(liner prediction cofficient),LPCC(liner prediction cepstrum coefficient) and LPCC(liner prediction cepstrum coefficient) characteristics from voice. Simulation results indicate that the detection rate reached 100% under the clean voice and more than 80% corresponding the noise level of -5 dB. This algorithm is effective under non-stationary nongaussian noise and the detection efficiency and performance are superior to other algorithms.
出处
《解放军理工大学学报(自然科学版)》
EI
2007年第1期10-14,共5页
Journal of PLA University of Science and Technology(Natural Science Edition)
基金
江苏省自然科学基金资助项目(BK2006001)
关键词
非高斯
语音激活检测
灰关联分析
特征
nongaussian
VAD(voice activity detection) gray correlation analysis
characteristics
作者简介
陈功(1979-)。男。博士生.
联系人:张雄伟。教授.博士生导师;研究方向:数字通信和多媒体信息处理等;E-mail:xwzhang@public.ptt.is.cn.