摘要
智能语音已经走进人们的日常生活,端点检测技术的发展对语音识别的应用起到了关键性作用,如何在嘈杂环境下正确识别出语音段,是影响语音识别系统性能的重要因素。针对已有的端点检测技术,提出了改进型多特征语音端点检测方法,在降噪的同时进行语音增强。即运用子带谱熵进行噪声估计,运用自适应噪声平滑进行降噪,并在谱减法的基础上进一步改进谱减参数,得到增强的语音信号。通过MATLAB仿真发现,改进型多特征语音端点检测技术能够更好地适应不同噪声环境,对端点检测有很好的适用性。
Intelligent voice has been increasingly going into the daily life,the development of endpoint detection technology plays a key role in the application of speech recognition,how to correctly identify speech segments in noisy environment has been becoming an important factor and affecting the effectiveness of a speech recognition system.Based on the original endpoint detection technology,an improved multi-feature speech endpoint detection method is proposed,which can enhance speech while reducing noise.The subband spectral entropy is used for noise estimation and adaptive noise smoothing is used for noise reduction.On the basis of spectral subtraction,the spectrum subtraction parameters are further improved to obtain the enhanced speech signal.Through MATLAB simulation,it is found that the improved multi-feature speech endpoint detection technology can better adapt to different noise environments,and has good applicability to endpoint detection.
作者
刘艳辉
LIU Yanhui(Department of Information and Media,Sanmenxia Polytechnic,Sanmenxia 472100,China)
出处
《河南工程学院学报(自然科学版)》
2022年第4期69-73,78,共6页
Journal of Henan University of Engineering:Natural Science Edition
基金
河南省高等学校重点科研项目(17A413010)
河南省科技攻关项目(182102210479)
河南省教育科学“十三五”规划项目(2020YB0632)
河南省高等职业学校青年骨干教师培养计划项目(2020GZGG041)。
关键词
语音识别
端点检测
谱减
MATLAB
speech recognition
endpoint detection
spectral subtraction
MATLAB