变噪声环境下的语音端点检测算法  

A Word Boundary Detection Algorithm for Variable Noise Environment

作　　者：何义华[1] 施玉霞[1] 

机构地区：[1]南京航空航天大学自动化学院,江苏南京210016

出　　处：《电声技术》2008年第8期64-67,71,共5页Audio Engineering

摘　　要：目前的语音端点检测算法大多不能适应背景噪声实时变化情况。针对这一问题,首先给出1个代表噪声变化的参数——最小Mel频带参数(MiMSB);然后给出改进的时频参数(TF)——增强时频参数(ETF),用于区分语音信号与噪声;最后基于这2个参数提出一种变噪声环境下的端点检测算法。实验证明,该算法在各种噪声环境下均能取得很好的性能。Many boundary detection algorithms, which focus only on the fixed background noise and high signalto-noise ratio, are proposed, but most of them cannot work well in the condition of the variable background noise. To solve this problem, MiMSB （Minimum Mel-Scale Frequency Band） parameter which can estimate the varying background noise level by adaptively choosing one band with minimum energy from the Mel-scale frequency band is proposed. Then, ETF（Enhanced Time-Frequency） parameter by extending TF（Time-Frequency） parameter from single band to multi-band spectrum analysis is proposed, where the frequency bands help to make the distinction between speech signal and noise. Based on the MiMSB and ETF parameters, a new algorithm for word boundary detection in variable noise environment is given. The new algorithm is tested over a variety of noise conditions and is found to perform well not only under variable background noise, but also under fixed background noise.

关 键 词：语音识别 端点检测 最小美尔频带参数 增强时频参数 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]