面向嵌入式系统多通道回声消除语音增强方法  

Multi channel echo cancellation and speech enhancement for embedded systems method

作　　者：张文杰 孙成立[1,2] ZHANG Wenjie;SUN Chengli(Key Laboratory of Image Processing and Pattern Recognition,Nanchang University of Aeronautics and Astronautics,Nanchang 330063;Nanchang University of Aeronautics and Astronautics School of Infomation Engineering,Nanchang 330063)

机构地区：[1]南昌航空大学图像处理与模式识别重点实验室,江西南昌330063 [2]南昌航空大学信息工程学院,江西南昌330063

出　　处：《长江信息通信》2024年第4期64-68,共5页Changjiang Information & Communications

摘　　要：语音增强在当代社会成为了一种语音前端处理技术,为了满足通信交流日益增长的需求,为了发掘效果更优、复杂度更低的语音增强系统,设计了一种面向嵌入式系统多通道回声消除语音增强方法,该方法结合了UNet网络结构和通道注意力机制,提出了两种回声消除算法模型:MP-UNet模型和轻量化版MP-UNetS模型,在两种模型中可以同时处理语音的幅值和相位,并输出消除回声和抑制噪声的语音。实验验证了所提方法中在性能上的优越性,在资源受限的嵌入式平台开发环境下对语音增强系统具有较大的意义。Speech enhanccmcnt has become a front-cnd proccssing technology in contemporary society.In order to meet the growing demand for communication and explore better and less complex speech enhancement systems,a multi-channel echo cancellation speech enhancement method for embedded systems has been designed,which combines the UNet network structure and channel attention mechanism,Two echo cancellation algorithm models have been proposed:the MP-UNet model and the lightweight version of the MP-UNets model.In both models,the amplitude and phase of specch can be processed simultancously,and speech that eliminates echocs and suppresses noise can be output.The experimental verification of the performance superiority of the proposed method has significant implications for speech enhancement systems in resource constrained embedded platform development environments.

关 键 词：嵌入式系统 深度学习 MP-UNet网络 语音增强 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]