期刊文献+
共找到72篇文章
< 1 2 4 >
每页显示 20 50 100
Research status and challenges in the manufacturing of IR conformal optics 被引量:2
1
作者 Jianbo Zhao Sheng Wang +2 位作者 Chunyu Zhang Jinhu Wang Qingliang Zhao 《Defence Technology(防务技术)》 SCIE EI CAS CSCD 2024年第10期154-172,共19页
The infrared conformal window is one of the most critical components in aircraft.Conformal windows with high performance bring low aberrations,high aerodynamic performance,reliability in extreme working environments,a... The infrared conformal window is one of the most critical components in aircraft.Conformal windows with high performance bring low aberrations,high aerodynamic performance,reliability in extreme working environments,and added value for aircraft.Through the past decades,remarkable advances have been achieved in manufacturing technologies for conformal windows,where the machining accuracy approaches the nanometer level,and the surface form becomes more complex.These advances are critical to aircraft development,and these manufacturing technologies also have significant reference values for other directions of the ultra-precision machining field.In this review,the infrared materials suitable for manufacturing conformal windows are introduced and compared with insights into their performances.The remarkable advances and concrete work accomplished by researchers are reviewed.The challenges in manufacturing conformal windows that should be faced in the future are discussed. 展开更多
关键词 conformal optics Window and dome Infrared material Ultra-precision grinding POLISHING Measurement
在线阅读 下载PDF
In-situ Growth of Conformal SnO_(2) Layers for Efficient Perovskite Solar Cells
2
作者 LIU Suolan LUAN Fuyuan +3 位作者 WU Zihua SHOU Chunhui XIE Huaqing YANG Songwang 《无机材料学报》 SCIE EI CAS CSCD 北大核心 2024年第12期1397-1403,I0009,共8页
Significant progress has recently been made in enhancing the power conversion efficiency(PCE)of perovskite solar cells(PSCs).The electron transport layer(ETL),as an essential component of PSCs,significantly influences... Significant progress has recently been made in enhancing the power conversion efficiency(PCE)of perovskite solar cells(PSCs).The electron transport layer(ETL),as an essential component of PSCs,significantly influences the performance of devices.Traditional spin-coating method for preparing the ETL fails to fully cover the cusp of FTO transparent conductive glass substrate,leading to direct contact between perovskite film and FTO substrate,which induces charge recombination and reduces the performance of PSCs.To address this issue,an in-situ growth method was proposed to prepare conformal SnO_(2) films on FTO glass substrates in this study.The resulting SnO_(2) films are not only dense and uniform,fully covering the cusp of the FTO glass substrates and reducing the contact area between the FTO substrates and the perovskite films,but also facilitating the formation of perovskite films with large grain sizes.Moreover,the conformal SnO_(2) films can improve the charge extraction at the SnO_(2)/perovskite interface,reduce the trap density and trap-assisted recombination in PSCs,and thus enhance the PCE of PSCs.Through comparative experiments,it is found that the PSCs with in-situ grown SnO_(2) films show an improved PCE of 21.97%,which significantly increased compared to that with spin-coated SnO_(2) films(20.93%).All above data demonstrate that the as-prepared SnO_(2) film can serve as an ideal ETL.It is worth mentioning that this method avoids the use of corrosive hydrochloric acid and toxic thioglycolic acid,and it can also be extended to ITO flexible transparent conductive substrates in the future. 展开更多
关键词 perovskite solar cell conformal SnO_(2)film in-situ growth
在线阅读 下载PDF
Low side lobe pattern synthesis using projection method with genetic algorithm for truncated cone conformal phased arrays 被引量:9
3
作者 Guoqi Zeng Siyin Li +1 位作者 Yan Zhang Shanwei L 《Journal of Systems Engineering and Electronics》 SCIE EI CSCD 2014年第4期554-559,共6页
A hybrid method for synthesizing antenna's three dimensional (3D) pattern is proposed to obtain the low sidelobe feature of truncated cone conformal phased arrays. In this method, the elements of truncated cone con... A hybrid method for synthesizing antenna's three dimensional (3D) pattern is proposed to obtain the low sidelobe feature of truncated cone conformal phased arrays. In this method, the elements of truncated cone conformal phased arrays are projected to the tangent plane in one generatrix of the truncated cone. Then two dimensional (2D) Chebyshev amplitude distribution optimization is respectively used in two mutual vertical directions of the tangent plane. According to the location of the elements, the excitation current amplitude distribution of each element on the conformal structure is derived reversely, then the excitation current amplitude is further optimized by using the genetic algorithm (GA). A truncated cone problem with 8x8 elements on it, and a 3D pattern desired side lobe level (SLL) up to 35 dB, is studied. By using the hybrid method, the optimal goal is accomplished with acceptable CPU time, which indicates that this hybrid method for the low sidelobe synthesis is feasible. 展开更多
关键词 conformal phased array low side lobe pattern synthe-sis projection method genetic algorithm optimization.
在线阅读 下载PDF
A novel DOA estimation algorithm using directional antennas in cylindrical conformal arrays 被引量:9
4
作者 Xiao-feng Gao Ping Li +2 位作者 Xin-hong Hao Guo-lin Li Zhi-jie Kong 《Defence Technology(防务技术)》 SCIE EI CAS CSCD 2021年第3期1042-1051,共10页
In this paper, a novel direction of arrival(DOA) estimation algorithm using directional antennas in cylindrical conformal arrays(CCAs) is proposed. To eliminate the shadow effect, we divide the CCAs into several subar... In this paper, a novel direction of arrival(DOA) estimation algorithm using directional antennas in cylindrical conformal arrays(CCAs) is proposed. To eliminate the shadow effect, we divide the CCAs into several subarrays to obtain the complete output vector. Considering the anisotropic radiation pattern of a CCA, which cannot be separated from the manifold matrix, an improved interpolation method is investigated to transform the directional subarray into omnidirectional virtual nested arrays without non-orthogonal perturbation on the noise vector. Then, the cross-correlation matrix(CCM) of the subarrays is used to generate the consecutive co-arrays without redundant elements and eliminate the noise vector. Finally, the full-rank equivalent covariance matrix is constructed using the output of co-arrays,and the unitary estimation of the signal parameters via rotational invariance techniques(ESPRIT) is performed on the equivalent covariance matrix to estimate the DOAs with low computational complexity. Numerical simulations verify the superior performance of the proposed algorithm, especially under a low signal-to-noise ratio(SNR) environment. 展开更多
关键词 Direction of arrival(DOA) conformal antenna array Interpolation technique Nested array
在线阅读 下载PDF
Conformal analysis of fundamental frequency of vibration of simple-supported elastic ellipse-plates with concentrated substance 被引量:2
5
作者 齐红元 朱衡君 《Journal of Central South University》 SCIE EI CAS 2005年第S2期269-273,共5页
Aimed at calculating the fundamental frequency of vibration of special-shaped, simple-supported elastic plates, Conformal Mapping theory is applied, and the mathematical method of trigonometric interpolation with inte... Aimed at calculating the fundamental frequency of vibration of special-shaped, simple-supported elastic plates, Conformal Mapping theory is applied, and the mathematical method of trigonometric interpolation with interpolation points mutual iterative between odd and even sequences in boundary region is provided, as well as the conformal mapping function which can be described by real number region between complicated region and unit dish region is carried out. Furthermore, in the in-plane state of constant stress, vibrating function is completed by unit dish region method for simple-supported elastic plates with concentrated substance of complicated vibrating region, and the coefficient of fundamental frequency of the plate is analyzed. Meanwhile, taking simple-supported elastic ellipse-plates as an example, the effects on fundamental frequency caused by eccentric ratio, the coefficient of constant in-plane stress, as well as the concentrated substance mass and positions are analyzed respectively. 展开更多
关键词 vibration fundamental frequency mode conformal mapping method of trigonometric interpolation elastic simple-supported plates
在线阅读 下载PDF
Direction of arrival estimation on cylindrical conformal array using RARE 被引量:1
6
作者 Kai Yang Zhiqin Zhao Wei Yang Zaiping Nie 《Journal of Systems Engineering and Electronics》 SCIE EI CSCD 2011年第5期767-772,共6页
When the information of mutual coupling and shadowing effect of a conformal antenna array are unknown, the performance of direction of arrival (DOA) estimation will be seriously degraded by using some classical meth... When the information of mutual coupling and shadowing effect of a conformal antenna array are unknown, the performance of direction of arrival (DOA) estimation will be seriously degraded by using some classical methods, such as the multiple signal classification (MUSIC) algorithm. Meanwhile it is difficult to measure or estimate the shadowing effect. The DOA estimation for a conformal uniform circular array (UCA) is studied. Firstly, the azimuthal angle is separated from all the unknown information by transforming the UCA from the element space to the mode space. Then the rank reduction (RARE) algorithm is applied in the estima- tion of the azimuthal angle. The π ambiguity existed in the RARE is solved by the beam forming. The main advantage of this method is that it does not need to measure the mutual coupling and the shadowing effect. Compared with the subarray method, it will not decrease the aperture of the array. Simulation results validate the advantages of the method. 展开更多
关键词 conformal antenna direction of arrival mutual coupling.
在线阅读 下载PDF
Joint polarization and DOA estimation based on improved maximum likelihood estimator and performance analysis for conformal array 被引量:1
7
作者 SUN Shili LIU Shuai +2 位作者 WANG Jun YAN Fenggang JIN Ming 《Journal of Systems Engineering and Electronics》 SCIE EI CSCD 2023年第6期1490-1500,共11页
The conformal array can make full use of the aperture,save space,meet the requirements of aerodynamics,and is sensitive to polarization information.It has broad application prospects in military,aerospace,and communic... The conformal array can make full use of the aperture,save space,meet the requirements of aerodynamics,and is sensitive to polarization information.It has broad application prospects in military,aerospace,and communication fields.The joint polarization and direction-of-arrival(DOA)estimation based on the conformal array and the theoretical analysis of its parameter estimation performance are the key factors to promote the engineering application of the conformal array.To solve these problems,this paper establishes the wave field signal model of the conformal array.Then,for the case of a single target,the cost function of the maximum likelihood(ML)estimator is rewritten with Rayleigh quotient from a problem of maximizing the ratio of quadratic forms into those of minimizing quadratic forms.On this basis,rapid parameter estimation is achieved with the idea of manifold separation technology(MST).Compared with the modified variable projection(MVP)algorithm,it reduces the computational complexity and improves the parameter estimation performance.Meanwhile,the MST is used to solve the partial derivative of the steering vector.Then,the theoretical performance of ML,the multiple signal classification(MUSIC)estimator and Cramer-Rao bound(CRB)based on the conformal array are derived respectively,which provides theoretical foundation for the engineering application of the conformal array.Finally,the simulation experiment verifies the effectiveness of the proposed method. 展开更多
关键词 conformal array maximum likelihood(ML)estimator manifold separation technology(MST) parameter estimation Cramer-Rao bound(CRB).
在线阅读 下载PDF
Conformal optimal design and processing of extruding die cavity
8
作者 齐红元 陈科山 杜凤山 《Journal of Central South University》 SCIE EI CAS 2008年第S2期357-361,共5页
Aimed at the optimal analysis and processing technology of die cavity of special-shaped products extrusion, by numerical analysis of trigonometric interpolation and Conformal Mapping theory, on the non-circle cross-se... Aimed at the optimal analysis and processing technology of die cavity of special-shaped products extrusion, by numerical analysis of trigonometric interpolation and Conformal Mapping theory, on the non-circle cross-section of special-shaped products, the conformal mapping function can be set up to translate the cross-section region into unit dish region, over numerical finite interpolation points between even and odd. Products extrusion forming can be turned into two-dimension problem, and plastic stream function can be deduced, as well as the mathematical model of the die cavity surface is established based on deferent kinds of vertical curve. By applying Upper-bound Principle, the vertical curves and related parameters of die cavity are optimized. Combining with electrical discharge machining (EDM) process and numerical control (NC) milling machine technology, the optimal processing of die cavity can be realized. Taking ellipse-shaped products as an instance, the optimal analysis and processing of die cavity including extruding experiment are carried out. 展开更多
关键词 EXTRUSION FORMING electrical DISCHARGE maching DIE cavity vertical CURVE conformal Mapping
在线阅读 下载PDF
Hybrid alternate projection algorithm and its application for practical conformal array pattern synthesis
9
作者 Fei Zhao Shunlian Chai +2 位作者 Huiying Qi Ke Xiao Junjie Mao 《Journal of Systems Engineering and Electronics》 SCIE EI CSCD 2012年第5期625-632,共8页
Based on the fabricated 12-element cavity-backed microstrip sector cylinder array,a novel hybrid alternate projection algorithm(HAPA),which combines analytical method with numerical techniques effectively,is propose... Based on the fabricated 12-element cavity-backed microstrip sector cylinder array,a novel hybrid alternate projection algorithm(HAPA),which combines analytical method with numerical techniques effectively,is proposed for synthesizing the pattern of practical conformal array.The algorithm applies the variable direction aperture projection method with mutual coupling correction techniques to provide the good initial excitations of elements to the enhanced alternate projection algorithm(EAPA).In order to do further optimization,which improves the convergent speed of the algorithm significantly.Finally,the HAPA has been applied to the fabricated sector cylinder array with mutual coupling considered.The results of synthesized patterns,such as low sidelobe with null points formed pattern,beam scanning with low sidelobe pattern and the shaped beam pattern are presented.It demonstrates the validity of HAPA in practical conformal array synthesis. 展开更多
关键词 conformal array alternate projection algorithm aperture projection method pattern synthesis mutual coupling
在线阅读 下载PDF
基于多特征迁移学习的低资源临高方言语音识别方法
10
作者 王忠 曹春杰 +3 位作者 谢夏 穆罕默德·艾哈迈德·拉扎 陈勇青 陈昱珏 《通信学报》 北大核心 2025年第10期221-232,共12页
针对低资源临高方言语音识别中数据稀缺、字错误率高的问题,提出了一种基于多特征迁移学习的端到端语音识别方法。以TeleSpeech-ASR1.0-large多方言预训练模型为基座,融合梅尔频率倒谱系数、滤波器组能量系数与对数梅尔谱3类互补声学特... 针对低资源临高方言语音识别中数据稀缺、字错误率高的问题,提出了一种基于多特征迁移学习的端到端语音识别方法。以TeleSpeech-ASR1.0-large多方言预训练模型为基座,融合梅尔频率倒谱系数、滤波器组能量系数与对数梅尔谱3类互补声学特征,通过构建Conformer-LAS-CTC联合优化架构,利用深度可分离卷积和多头自注意力机制分别捕捉语音信号的局部特征与全局依赖关系,并设计融合CTC、中间层CTC与注意力机制的多任务损失函数进行联合训练。在总时长为280 h的临高方言与普通话混合语料上的实验结果表明,所提方法的字错误率降低至18.89%,显著优于基线模型,有效缓解了低资源方言面临的数据瓶颈问题,为濒危语言的数字化保护提供了可行的技术路径。 展开更多
关键词 低资源语音识别 迁移学习 CONFORMER 多特征融合 临高方言
在线阅读 下载PDF
吉林大学杨雨欣论文被IEEE S&P 2026接收
11
《信息网络安全》 北大核心 2025年第10期1641-1641,共1页
吉林大学计算机科学与技术学院2022级博士研究生杨雨欣为第一作者的论文“Ensemble Conformal Predictor (En CP):A New Conformal Predictor with Robustness Guarantees Against Data Poisoning Attacks”被IEEE Symposium on Securit... 吉林大学计算机科学与技术学院2022级博士研究生杨雨欣为第一作者的论文“Ensemble Conformal Predictor (En CP):A New Conformal Predictor with Robustness Guarantees Against Data Poisoning Attacks”被IEEE Symposium on Security and Privacy (IEEE S&P 2026)接收。作者还包括杨雨欣的指导教师教授李强、吉林大学人工智能学院博士研究生封润洋,共同通信作者是美国丰田工业大学芝加哥分校教授Liren Shan和美国伊利诺伊理工大学教授Binghui Wang。 展开更多
关键词 En CP Ensemble conformal Predictor
在线阅读 下载PDF
基于语音信号时频特征融合的帕金森病检测方法 被引量:1
12
作者 王晨哲 季薇 +1 位作者 郑慧芬 李云 《郑州大学学报(理学版)》 CAS 北大核心 2025年第1期53-60,共8页
发音障碍是帕金森病的早期症状之一。近年来,基于语音信号的帕金森病检测的研究大多采用梅尔刻度下的相关语音特征与深度神经网络模型相结合的方法。然而,现有的模型无法充分关注语音信号的全局时序信息,且梅尔刻度特征在准确表征帕金... 发音障碍是帕金森病的早期症状之一。近年来,基于语音信号的帕金森病检测的研究大多采用梅尔刻度下的相关语音特征与深度神经网络模型相结合的方法。然而,现有的模型无法充分关注语音信号的全局时序信息,且梅尔刻度特征在准确表征帕金森病的病理信息方面效果有限。为此,提出了一种基于语音时频特征融合的帕金森病检测方法。首先,提取语音的梅尔频率倒谱系数,并将其作为模型的输入。接着,在已有的S-vectors模型中引入Conformer编码器模块,以提取语音的时域全局特征。最后,将与帕金森病语音检测相关的频域全局特征嵌入时域特征中进行时频信息融合,以实现帕金森病语音检测。在公开帕金森病语音数据集和自采语音数据集上验证了方法的有效性。 展开更多
关键词 帕金森病 梅尔频率倒谱系数 S-vectors CONFORMER 时频特征融合
在线阅读 下载PDF
改进Transformer解码器的端到端语音识别 被引量:1
13
作者 胡恒博 牛铜 何振华 《计算机应用》 北大核心 2025年第S1期95-100,共6页
Transformer模型架构在序列到序列任务中可以很好地将注意力分散到整个输入上以学习长期依赖关系,然而,在语音识别中,文本输出和语音输入是单调对齐的。针对Transformer解码器无法较好地捕获局部特征以进行单调对齐的问题,提出一种改进... Transformer模型架构在序列到序列任务中可以很好地将注意力分散到整个输入上以学习长期依赖关系,然而,在语音识别中,文本输出和语音输入是单调对齐的。针对Transformer解码器无法较好地捕获局部特征以进行单调对齐的问题,提出一种改进的Transformer解码器。将Transformer解码器中的2种注意力机制拆分为2个单独模块,再使用交叉注意力进行更高效的局部特征捕获。在开源中文普通话AISHELL-1数据集上的实验结果表明,使用能够捕获局部特征的编码器时,该解码器相较于Transformer解码器有着更好的识别效果。具体地,当编码器为Conformer时,字错误率(CER)降低了16.19%,且收敛速度更快,而在使用了连接时序分类(CTC)进行辅助解码后,CER降低了5.08%,最终的CER为4.67%。 展开更多
关键词 交叉注意力 Transformer解码器 Conformer编码器 语音识别 局部特征
在线阅读 下载PDF
融合双通道卷积和改进型Conformer的两阶段语音增强算法
14
作者 徐佳瑜 郑展恒 +1 位作者 曾庆宁 王健 《电子测量技术》 北大核心 2025年第4期149-157,共9页
针对语音关键特征提取不充分、模型结构单一的问题,提出一种两阶段下融合多尺度特征和改进型门控Conformer的语音增强方法。首先,针对关键特征提取不充分的问题,提出双通道卷积融合模块,采用不同感受野的二维卷积多尺度提取语音关键信息... 针对语音关键特征提取不充分、模型结构单一的问题,提出一种两阶段下融合多尺度特征和改进型门控Conformer的语音增强方法。首先,针对关键特征提取不充分的问题,提出双通道卷积融合模块,采用不同感受野的二维卷积多尺度提取语音关键信息,并结合门控机制增强网络的短期与长期序列相关性,从而提升模型在复杂环境下的语音增强效果;提出改进型Conformer,采用时间注意和频率注意分别在时域和频域上进行建模,并结合膨胀卷积模块高效提取局部与全局上下文信息,从而增强网络在语音序列建模中的表现能力。其次,针对模型结构单一的问题,采用两阶段处理结构,将复杂问题分步处理。在第一阶段首先接收噪声频谱的幅值,初步估计出干净语音的幅值,并与噪声相位进行重构,得到粗糙的复频谱。第二阶段在第一阶段得到粗谱的基础上进一步提取更精细的特征,增强语音信号的细节表现能力。最后,在VoiceBank+DEMAND数据集上进行测试,实验结果表明,所提算法相比带噪语音的语音感知质量和短时客观可懂度分别提升50.25%、3.26%,表明该网络能够更有效地提高语音的可懂度,同时改善语音信号的整体质量,具有较强的降噪能力。 展开更多
关键词 深度学习 语音增强 CONFORMER 多尺度特征提取 两阶段
在线阅读 下载PDF
结合字节级别字节对编码的端到端中文语音识别方法
15
作者 付强 徐振平 +1 位作者 盛文星 叶青 《计算机应用》 北大核心 2025年第1期318-324,共7页
针对语音识别中对中文这种复杂字符集的语言词汇表过大以及训练效率太低的问题,提出一种基于字节级别字节对编码(BBPE)的端到端中文语音识别方法。首先,将256个不同的字节用于初始化词汇表;其次,统计每个词汇单元在语料中出现的频率,并... 针对语音识别中对中文这种复杂字符集的语言词汇表过大以及训练效率太低的问题,提出一种基于字节级别字节对编码(BBPE)的端到端中文语音识别方法。首先,将256个不同的字节用于初始化词汇表;其次,统计每个词汇单元在语料中出现的频率,并合并频率最高的词汇单元;最后,重复上一步直至无法合并,以得到最终的词汇表。在中文语音数据集AISHELL-1上,该方法生成的词汇表相较于字符级别词汇表的词汇量减少了88.5%,降低了模型训练的复杂度。同时,鉴于Conformer-Transducer(Conformer-T)模型在端到端语音识别中的出色表现,为了实现更好的识别效果,将最新的Zipformer模型与Transducer模型相结合提出Zipformer-Transducer(Zipformer-T)模型,并在该模型上对BBPE方法进行验证。实验结果表明,Zipformer-T模型使用的BBPE方法相较于字符级别分词方法在AISHELL-1测试集和验证集上的字错率(CER)分别降低了0.12和0.08个百分点,且分别达到4.26%和3.98%的最低CER,充分说明该方法能有效提升中文语音识别的性能。 展开更多
关键词 语音识别 CONFORMER Zipformer 字节级别字节对编码 端到端
在线阅读 下载PDF
基于Conformer-LSTM模型的连续无创血压预测方法
16
作者 陈欣 刘立程 王小林 《电子测量技术》 北大核心 2025年第15期120-128,共9页
本研究提出了一种基于Conformer-LSTM模型的连续无创血压预测方法,模型包括卷积支路、Transformer支路、两个多尺度交叉注意力模块、自适应空间特征融合模块和两层LSTM。通过该方法,仅通过输入PPG信号即可预测对应的ABP波形,收缩压和舒... 本研究提出了一种基于Conformer-LSTM模型的连续无创血压预测方法,模型包括卷积支路、Transformer支路、两个多尺度交叉注意力模块、自适应空间特征融合模块和两层LSTM。通过该方法,仅通过输入PPG信号即可预测对应的ABP波形,收缩压和舒张压通过预测的ABP波形得出。此外,该方法在较大的数据集中取得较小的预测误差,实验结果表明,本文提出的模型在MIMIC数据集中预测的ABP波形与实际波形的拟合程度较好,SBP和DBP的预测误差分别为(3.68±5.60)mmHg和(2.16±3.72)mmHg,该方法符合美国医疗仪器促进协会(AAMI)标准,并在英国高血压协会(BHS)标准中获得A级评价。 展开更多
关键词 血压预测 多尺度特征融合 CONFORMER PPG信号
在线阅读 下载PDF
多任务学习型民航陆空通话语音识别Conformer模型
17
作者 马广林 任晋 +3 位作者 师一华 张海刚 王莉 杨金锋 《计算机应用与软件》 北大核心 2025年第10期183-190,244,共9页
民航陆空通话在用语发音、遣词造句和通话方式等方面具有显著行业特点,通用语音识别模型无法充分适配上述特点对陆空通话进行声学建模。针对上述问题,提出一种端到端的多任务学习型民航陆空通话语音识别Conformer模型。通过将卷积模块引... 民航陆空通话在用语发音、遣词造句和通话方式等方面具有显著行业特点,通用语音识别模型无法充分适配上述特点对陆空通话进行声学建模。针对上述问题,提出一种端到端的多任务学习型民航陆空通话语音识别Conformer模型。通过将卷积模块引入Transformer模型,Conformer模型在保留上下文长距离依赖关系的全局信息建模能力基础上,进一步增强了局部信息的捕获。同时联合连接时序分类(Connectionist Temporal Classification,CTC)和基于注意力的编码解码模型进行多任务学习以进一步提升其性能。实验结果表明,该方法能有效兼顾全局和局部信息的声学建模,在陆空通话数据集上将字符错误率和句错误率分别降低至1.98%和2.89%。 展开更多
关键词 民航陆空通话 语音识别 多任务学习 CONFORMER 端到端
在线阅读 下载PDF
ACGFN:基于非对称卷积和门控前馈神经网络的语音识别模型 被引量:1
18
作者 王詠森 刘倩 刘立波 《中文信息学报》 北大核心 2025年第1期167-174,共8页
针对现有基于Conformer语音识别模型对时频特征提取能力不足、模型结构冗余和参数量较大的问题,该文提出一个基于非对称卷积和门控前馈神经网络的语音识别模型ACGFN。首先,采用不同感受野大小的非对称卷积对语音序列的时频特征进行多尺... 针对现有基于Conformer语音识别模型对时频特征提取能力不足、模型结构冗余和参数量较大的问题,该文提出一个基于非对称卷积和门控前馈神经网络的语音识别模型ACGFN。首先,采用不同感受野大小的非对称卷积对语音序列的时频特征进行多尺度融合下采样,在增强模型提取时频特征的能力的同时,有效降低了下采样过程中信息的损失;其次,引入门控前馈模块替换Conformer中的双半步前馈网络,降低网络参数量的同时精简了模型结构。实验结果表明,该方法在公共数据集AISHELL-1和aidatatang_200zh的测试集上字错误率分别为4.48%、4.28%,且参数量仅40.3M。相较对比方法,识别字错误率和参数量均有所降低。 展开更多
关键词 语音识别 端到端 CONFORMER
在线阅读 下载PDF
基于多视角注意力的异构双分支解码单通道语音增强
19
作者 更藏措毛 黄鹤鸣 《计算机应用》 北大核心 2025年第10期3284-3293,共10页
针对单通道语音增强中主流编解码结构面临的声学特征提取不充分、通道信息丢失和幅度相位补偿困难等问题,提出一种融合不同维度语音特征的异构双分支解码单通道语音增强模型——HDBMV(Heterogeneous DualBranch with Multi-View)。该模... 针对单通道语音增强中主流编解码结构面临的声学特征提取不充分、通道信息丢失和幅度相位补偿困难等问题,提出一种融合不同维度语音特征的异构双分支解码单通道语音增强模型——HDBMV(Heterogeneous DualBranch with Multi-View)。该模型通过信息融合编码器(IFE)、时频残差Conformer(TFRC)模块、多视角注意力(MVA)模块和异构双分支解码器(HDBD)等机制,提升单通道语音增强的性能。首先,IFE联合处理振幅与复数特征,捕捉全局依赖和局部相关,生成紧凑的特征表示;其次,TFRC模块有效捕捉时间维度和频域维度上的相关性,同时降低计算复杂度;再次,MVA模块重构通道域和时频域信息,进一步增强模型对信息的多视角多层次的表征能力;最后,HDBD分别处理幅度特征和细化复数特征,解决幅度相位补偿问题,提升解码鲁棒性。实验结果表明,HDBMV在公开数据集VoiceBank+DEMAND、大数据集DNS Challenge 2020和自建的藏语数据集BodSpeDB上的语音质量感知评估(PESQ)分别达到了3.00、3.12和2.09,短时目标可理解度(STOI)分别达到了0.96、0.97和0.81。可见,HDBMV以最小的参数量和较高的计算效率获得了最佳的语音增强性能和较强的泛化能力。 展开更多
关键词 语音增强 编解码器 CONFORMER 注意力机制 复数特征
在线阅读 下载PDF
WTSTC:基于广域时频采样和时序感知卷积的语音识别模型
20
作者 刘立波 王詠森 +1 位作者 刘倩 邓箴 《中文信息学报》 北大核心 2025年第4期161-171,共11页
针对现有语音识别模型存在的时频特征感受野不足、时序特征损失及模型结构扩展性较差等方面的问题,该文提出基于广域时频采样和时序感知卷积的语音识别模型WTSTC,在保证模型轻量化的同时提升识别精度。首先,通过结合RepLKNet模块和传统... 针对现有语音识别模型存在的时频特征感受野不足、时序特征损失及模型结构扩展性较差等方面的问题,该文提出基于广域时频采样和时序感知卷积的语音识别模型WTSTC,在保证模型轻量化的同时提升识别精度。首先,通过结合RepLKNet模块和传统卷积下采样模块,构建了一种新型的广域时频采样模块,增大感受野的同时更加关注输入音频序列的时频特征;其次,设计了时序感知卷积模块,通过实现应用于时序特征的一维全局响应归一化层取代原有的Batch Norm以增强通道间的特征竞争,避免了归一化过程中语音信号的时序特征信息丢失的潜在可能;最后,在模型内部各模块间引入Droppath正则化方法,通过在模块间随机跳跃样本避免模型对特定模块的依赖。实验结果表明,该方法在中文公共数据集AISHELL-1的测试集上字错率为4.27%,在更大规模英文公共数据集Librispeech的测试集clean和other上的词错率分别为2.2%和5.1%。在保持相同训练策略的前提下,该方法相较现有先进模型展现出更优异的性能。 展开更多
关键词 自动语音识别 端到端 CONFORMER RepLKNet
在线阅读 下载PDF
上一页 1 2 4 下一页 到第
使用帮助 返回顶部