语音增强降噪方法研究  

Research on Speech Enhancement and Noise Reduction Method

作　　者：孟飞宇 张红兵[1] 岳沛瑀 MENG Feiyu;ZHANG Hongbing;YUE Peiyu(School of Information Technology and Intelligence of Public Security,Criminal Investigation Police University of China,Shenyang Liaoning 110854,China)

机构地区：[1]中国刑事警察学院公安信息技术与情报学院,辽宁沈阳110854

出　　处：《辽宁警察学院学报》2023年第2期78-81,共4页Journal of Liaoning Police College

基　　金：中国刑事警察学院研究生创新能力提升项目“基于云计算平台的语音信号处理技术应用研究”(2021YCYB46)。

摘　　要：语音信号在采集和传输过程中会受到噪声的干扰,造成音频信噪比过低。通常会用谱减法、滤波器法、神经网络等方法对收集到的音频进行降噪处理。本研究以模型算法和非模型算法两部分为研究顺序,分析了几种常见降噪方法的原理和优缺点,通过对相应代码的实现,以信噪比和分段信噪比为评价标准,对比分析各类方法的降噪效果,最后得出基于LMS自适应滤波器降噪方法效果更明显,同时神经网络的语音降噪方法会有更好的应用前景。The voice signal is interfered by noise in the process of acquisition and transmission,resulting in low signal to noise ratio of audio.Usually,spectral subtraction,filter method,neural network and other methods are used to denoise the collected audio.This paper takes model algorithm and non-model algorithm as the research order,analyzes the principle,advantages and disadvantages of several common noise reduction methods,and compares and analyzes the noise reduction effects of various methods through the implementation of corresponding codes,taking SNR and segmented SNR as the evaluation criteria.Finally,it is concluded that the noise reduction method based on LMS adaptive filter is more effective,and the voice noise reduction method based on neural network will have better application prospects.

关 键 词：降噪 谱减法 滤波器法 神经网络 

分 类 号：D918.2[政治法律—法学]