摘要
为了提升模型在复杂场景下的信息处理能力,提出了一种基于注意力的动态选择机制,根据当前信息选择性地分配权重,有效融合形变卷积和普通卷积的特征输出,自适应地在卷积形变和标准卷积之间进行权衡,从而提高其表示能力。此外,通过借鉴渐进学习,在不增加额外参数的前提下,通过循环迭代的方式进一步增强了模型的学习能力。在TIMIT公开语料库上使用七种来自NoiseX92的不同噪声,在多种信噪比环境下进行实验,结果表明无论信噪比高低,噪声是否在训练数据集中出现,所提出的算法在可懂度和语音质量等客观评价指标上均优于近期其他的深度学习算法。
In order to improve the information processing ability of the model in complex scenes,this paper proposed a dynamic selection mechanism based on attention,which selectively allocated weights according to the current information,effectively fused the feature outputs of deformation convolution and ordinary convolution,and adaptively balanced deformable convolution and standard convolution,so as to improve its representation ability.In addition,the learning ability of the model was further enhanced by means of iteration without additional parameters.It used seven different kinds of noises from Noise-X92 in TIMIT corpus,and carried out experiments in various SNR environments.The results show that the proposed algorithm outperforms other recent deep learning algorithms in terms of intelligibility and speech quality,regardless of SNR and whether noise appears in the training data set.
作者
台文鑫
王钇翔
李森
蓝天
刘峤
Tai Wenxin;Wang Yixiang;Li Sen;Lan Tian;Liu Qiao(School of Information&Software Engineering,University of Electronic Science&Technology of China,Chengdu 610054,China)
出处
《计算机应用研究》
CSCD
北大核心
2021年第9期2604-2608,共5页
Application Research of Computers
基金
国家自然科学基金项目(U19B2028,61772117)
科技委创新特区项目(19-163-21-TS-001-042-01)
提升政府治理能力大数据应用技术国家工程实验室重点项目(10-2018039)
中央高校基本科研业务费项目(ZYGX2019J077)。
关键词
语音增强
低信噪比
动态选择机制
形变卷积
渐进学习
speech enhancement
low SNR
dynamic selection
deformable convolution
progressive learning
