一种基于LPC距离的端点检测改进算法  

An Improved Voice Activity Detection Method Based on Linear Predictive Coding Distance

作　　者：魏峰[1] 徐成[1] 赵景远[1] 

机构地区：[1]湖南大学计算机与通信学院,长沙410082

出　　处：《微处理机》2007年第4期46-48,52,共4页Microprocessors

基　　金：湖南省自然科学基金项目(03JJY3103);湖南省自然科学基金计划项目(05FJ3018)

摘　　要：线性预测编码系数是能反应人类语音信号的重要特征,在安静环境下或信噪比较高的情况下,基于线性预测编码系数可以对语音信号进行准确的端点检测和语音识别。针对嵌入式设备资源和成本有限的情况,提出了一种基于线性预测编码系数距离的端点检测改进算法,在分帧阶段,引入动态调整窗长的方法,减少计算量,在语音和非语音的过渡段使用较小的窗长,增加检测的准确度。实验结果证明,改进后的算法能更加准确的区分语音和非语音段,更适合应用于噪声较强的情况下。Linear predictive coding coefficient is a important character which can reflect human＇s speech signal. In a quiet or high Signal -to -Noise condition, it can be used to recognise speech signal and endpoint detective exactly. Aiming at the limit of embedded set$ resource and cost, proposes an improved endpoint detection algorithm based on linear predictive coding coefficient, brings in a dynamic length adjustment method in frame division to decrease the calculation. In the transition section between speech and non - speech, uses shorter window length to increase the accuracy. The experiment result shows that the improved algorithm can distinguish the speech and non - speech segments more accurately, it is more adapted to high noise condition.

关 键 词：端点检测 线性预测编码系数 动态窗长 

分 类 号：TN912.34[电子电信—通信与信息系统]