摘要
当今人工智能发展迅速,语音识别成为人机交互的重要方式。为提高语音识别准确度,在分析语音信号前去除语音信号噪声干扰并提高语音信号能量尤为重要。在实际应用中,不同语音信号包含不同的噪声。针对不同的语音噪声,在传统谱减法基础上,通过判断算法窗函数,根据不同的噪声能量改变多窗谱减法的过减因子参数,以增强算法自适应能力。仿真结果表明,在低信噪比情况下,通过改变过减因子值,可取得一个最优过减因子值以改进谱减法下的音乐噪声和失真度。自适应多窗谱减法改进后与基本谱减法相比,信噪比提高了29%;与多窗谱减法相比,信噪比提高了16%。该自适应多窗谱减法可适应不同噪声环境下的语音信号,增强语音信号中的关键信息并减少噪声干扰。
With the rapid development of artificial intelligence,speech recognition has become an important way of human-computer interaction.In order to improve the accuracy of speech recognition,it is particularly important to remove the noise interference of the speech signal and improve the speech signal energy before analyzing the speech signal.In practical applications,different speech sig⁃nals contain different noises.For the different speech noises,based on the traditional spectral subtraction method,the adaptive func⁃tion of the algorithm is enhanced by judging the window function in the algorithm and changing the over-subtraction factor parameters of the multi-window spectrum subtraction according to different noise energies.The simulation results show that in the case of low SNR,by changing the value of the over-subtraction factor,an optimal over-subtraction factor value is obtained to improve the music noise and distortion under spectral subtraction.The improved adaptive multi-window spectral subtraction has a 29%increase in sig⁃nal-to-noise ratio compared to the basic spectral subtraction,and a 16%increase in signal-to-noise ratio compared to multi-window spectral subtraction.The improved adaptive multi-window spectrum subtraction can adapt to speech signals in different noise environ⁃ments,enhance key information in the speech signal and reduce noise interference.
作者
张莉
李文钧
岳克强
ZHANG Li;LI Wen-jun;YUE Ke-qiang(Electronic Information School,Hangzhou Dianzi University,Hangzhou 310018,China)
出处
《软件导刊》
2020年第5期74-77,共4页
Software Guide
基金
浙江省重点研发计划项目(2019C03088)。
关键词
谱减法
自适应参数
多窗函数
语音降噪
spectral subtraction
adaptive parameters
multi-window function
speech noise reduction
