为了改善基于卷积编解码架构的单通道语音增强网络对语音声学特征提取不充分、解码特征丢失严重的问题,提出一种基于多路信息聚合协同解码的单通道语音增强网络MIACD,通过双路编码器充分提取融入了语音自监督学习(SSL)表征的幅度谱和复...为了改善基于卷积编解码架构的单通道语音增强网络对语音声学特征提取不充分、解码特征丢失严重的问题,提出一种基于多路信息聚合协同解码的单通道语音增强网络MIACD,通过双路编码器充分提取融入了语音自监督学习(SSL)表征的幅度谱和复数谱特征,由4层Conformer分别从时间和频率维度对提取特征建模,采用残差连接将双路编码器提取的语音幅度、复数特征引入三路信息聚合解码器,并利用所提通道-时频注意力(CTF-Attention)机制根据语音能量分布情况调节解码器中聚合信息,有效缓解解码时可用声学信息缺失严重的问题。在公开数据集Voice Bank DEMAND上的实验结果表明,与用于单通道语音增强的协作学习框架(GaGNet)相比,MIACD在客观评价指标宽带感知评估语音质量(WB-PESQ)上提升了5.1%,短时客观可懂度(STOI)达到96.7%,验证所提方法可充分利用语音信息重构信号,有效抑制噪声并提升语音可理解性。展开更多
In order to improve the performance of organic luminescent materials,lots of studies have been carried out at the molecular level.However,these materials are mostly applied as solids or aggregates in practical applica...In order to improve the performance of organic luminescent materials,lots of studies have been carried out at the molecular level.However,these materials are mostly applied as solids or aggregates in practical applications,in which the relationship between aggregation structure and luminescent property should be paid more attention.Here,we obtained five phenothiazine 5,5-dioxide(O-PTZ)derivatives with distinct molecular conformations by rational design of chemical structures,and systematically studied their room-temperature phosphorescence(RTP)effect in solid state.It was found that O-PTZ dimers with quasi-equatorial(eq)conformation tended to show strongerπ-πinteraction than quasi-axial(ax)conformers in crystal state,which was more conducive to the generation of RTP.Based on this result,a multi-level structural model of organic solids was proposed to draw the relationship between aggregation structure and RTP effect,just like the research for the structureproperty relationship of proteins.Using this structural model as the guide,boosted RTP efficiency from 1%to 20%was successfully achieved in the corresponding host-vip doping system,showing its wide applicability.展开更多
MIME(Multipurpose Internet Mail Extensions)主要用于电子邮件传输非文本数据。它兼容旧版本的信息格式,使得那些旧的应用网关也可以处理MIME格式的信息,并增加了许多功能,可以用邮件传输二进制数据。集合文档是MIME格式的一种应用,...MIME(Multipurpose Internet Mail Extensions)主要用于电子邮件传输非文本数据。它兼容旧版本的信息格式,使得那些旧的应用网关也可以处理MIME格式的信息,并增加了许多功能,可以用邮件传输二进制数据。集合文档是MIME格式的一种应用,用于将根资源和附属资源集合进同一个信息里。文章首先介绍了MIME格式和集合文档的格式,给出了一个对MIME格式文档的解码算法,然后在这个算法的基础上给出了集合文档的解码算法。展开更多
针对RGB-D(Red Green Blue Depth)语义分割中色彩信息和深度信息无法有效融合以及无法充分提取多尺度上下文信息的问题,文中提出了一种基于双流聚合Transformer的RGB-D语义分割方法。通过Transformer提取全彩图像和深度图像的多层次特征...针对RGB-D(Red Green Blue Depth)语义分割中色彩信息和深度信息无法有效融合以及无法充分提取多尺度上下文信息的问题,文中提出了一种基于双流聚合Transformer的RGB-D语义分割方法。通过Transformer提取全彩图像和深度图像的多层次特征,采用通道注意交叉融合模块与深度增强RGB操作实现各层次特征模态鸿沟的补偿,完成双模态信息融合。使用多层聚合解码器模块整合多层次多尺度上下文特征,减少了信息传递损失,实现了更准确和更全面的语义分割。实验结果表明,所提方法在NYU-Dv2数据集上的平均交并比(mean Intersection over Union,mIoU)、像素准确率和平均像素准确率分别达到52.9%、78.0%、66.0%。在Cityscapes数据集上的实验结果表明,在低分辨率输入图像下,所提方法的mIoU达到了79.8%。展开更多
文摘为了改善基于卷积编解码架构的单通道语音增强网络对语音声学特征提取不充分、解码特征丢失严重的问题,提出一种基于多路信息聚合协同解码的单通道语音增强网络MIACD,通过双路编码器充分提取融入了语音自监督学习(SSL)表征的幅度谱和复数谱特征,由4层Conformer分别从时间和频率维度对提取特征建模,采用残差连接将双路编码器提取的语音幅度、复数特征引入三路信息聚合解码器,并利用所提通道-时频注意力(CTF-Attention)机制根据语音能量分布情况调节解码器中聚合信息,有效缓解解码时可用声学信息缺失严重的问题。在公开数据集Voice Bank DEMAND上的实验结果表明,与用于单通道语音增强的协作学习框架(GaGNet)相比,MIACD在客观评价指标宽带感知评估语音质量(WB-PESQ)上提升了5.1%,短时客观可懂度(STOI)达到96.7%,验证所提方法可充分利用语音信息重构信号,有效抑制噪声并提升语音可理解性。
基金National Natural Science Foundation of China,Grant/Award Numbers:52273191,22235006Natural Science Foundation of Tianjin City,Grant/Award Number:22JCYBJC00760+3 种基金Open Project Program of Wuhan National Laboratory for Optoelectronics,Grant/Award Number:2020WNLOKF013starting Grants of Tianjin University and Tianjin GovernmentIndependent Innovation Fund of Tianjin University,Grant/Award Number:2023XPD-0014Guangzhou AIE Higher Research Institute。
文摘In order to improve the performance of organic luminescent materials,lots of studies have been carried out at the molecular level.However,these materials are mostly applied as solids or aggregates in practical applications,in which the relationship between aggregation structure and luminescent property should be paid more attention.Here,we obtained five phenothiazine 5,5-dioxide(O-PTZ)derivatives with distinct molecular conformations by rational design of chemical structures,and systematically studied their room-temperature phosphorescence(RTP)effect in solid state.It was found that O-PTZ dimers with quasi-equatorial(eq)conformation tended to show strongerπ-πinteraction than quasi-axial(ax)conformers in crystal state,which was more conducive to the generation of RTP.Based on this result,a multi-level structural model of organic solids was proposed to draw the relationship between aggregation structure and RTP effect,just like the research for the structureproperty relationship of proteins.Using this structural model as the guide,boosted RTP efficiency from 1%to 20%was successfully achieved in the corresponding host-vip doping system,showing its wide applicability.
文摘MIME(Multipurpose Internet Mail Extensions)主要用于电子邮件传输非文本数据。它兼容旧版本的信息格式,使得那些旧的应用网关也可以处理MIME格式的信息,并增加了许多功能,可以用邮件传输二进制数据。集合文档是MIME格式的一种应用,用于将根资源和附属资源集合进同一个信息里。文章首先介绍了MIME格式和集合文档的格式,给出了一个对MIME格式文档的解码算法,然后在这个算法的基础上给出了集合文档的解码算法。
文摘针对RGB-D(Red Green Blue Depth)语义分割中色彩信息和深度信息无法有效融合以及无法充分提取多尺度上下文信息的问题,文中提出了一种基于双流聚合Transformer的RGB-D语义分割方法。通过Transformer提取全彩图像和深度图像的多层次特征,采用通道注意交叉融合模块与深度增强RGB操作实现各层次特征模态鸿沟的补偿,完成双模态信息融合。使用多层聚合解码器模块整合多层次多尺度上下文特征,减少了信息传递损失,实现了更准确和更全面的语义分割。实验结果表明,所提方法在NYU-Dv2数据集上的平均交并比(mean Intersection over Union,mIoU)、像素准确率和平均像素准确率分别达到52.9%、78.0%、66.0%。在Cityscapes数据集上的实验结果表明,在低分辨率输入图像下,所提方法的mIoU达到了79.8%。
