面向噪声环境下医疗语音信号端点检测方法  

A Method for Detecting Endpoints of Medical Speech Signal in Specific Noise Environment

作　　者：刘永俊[1] 张立飞[1] 刘巍 

机构地区：[1]常熟理工学院计算机科学与工程学院,江苏常熟215500 [2]北京医优米信息技术有限公司,北京100089

出　　处：《常熟理工学院学报》2017年第4期75-79,85,共6页Journal of Changshu Institute of Technology

基　　金：江苏省高校自然科学基金项目"微纳米光学图像特征分析及多维度测量"(15KJB520001)

摘　　要：在语音信号处理领域,端点检测作为语音信号处理的初始化工作,有效合理的端点检测方法不仅可以减少数据的存储量和处理时间,而且可以排除无声段的噪声干扰,使语音的后续处理更加准确.本文为医疗语音信号处理提供一种医院噪声环境下医疗语音信号的端点检测方法.综合采用频域下的特征倒谱距离、能量谱、能量谱方差、统计学特征谱熵值4种特征作为控制条件判断语音的端点.通过设计软件验证了算法的有效性.In the field of speech signal processing, endpoint detection as the initialization of voice signal processing, effective and reasonable endpoint detection method can not only reduce the data storage capacity and processing time, but it can also eliminate the noise of the noise segment, so that the subsequent processing of voice can become more accurate. A method of detecting the endpoint of medical voice signal in noise environment is proposed, which provides an accurate endpoint detection method for medical voice signal processing in hospital noise environment. The endpoints of speech are judged by four characteristics including crosstalk distances, energy spectrum, energy spectrum variance and statistical characteristic spectral entropy. The effectiveness of the algorithm is verifed by the design software.

关 键 词：语音信号处理 语音端点检测 语音特征 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]