检索结果-内蒙古大学图书馆

6th International Symposium on Computer and Information processing Technology, ISCIPT 2021

作者： Cao, Zhonghui Huang, Zhihua Xinjaing University College of Information Science and Engineering Key Laboratory of Signal Detection and Processing in Xinjiang Uygur Autonomous Region Urumqi830001 China

ISBN: (纸本)9781665441377

Generative adversarial networks (GAN) achieve a good level of success in the speech enhancement (SE) task in recent years. However, traditional GAN for SE generates enhanced speech only as a function of local points, such that the model has difficulity in capturing long-range dependency and non-local information. In this paper, a model named self-attention GAN (ASEGAN) is proposed to estabilish connections between every sample and the rest samples in the time domain. Attention mechanism is used to model long-range dependency modeling for SE tasks. Armed with self-attention, the generator can generate the enhanced speech in which fine details at every sample are carefully coordinated with fine details in distant portions of the speech. Moreover, the discriminator can also more accurately enforce constraints on the global speech structure, thus the quality of enhanced speech by generator is better. Experimental results demonstrate that the proposed approach outperforms the competing methods measured by objective quality and speech intelligibility. © 2021 IEEE.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

RFBiCF: A Relation-First Bidirectional Cascade Framework for Relational Triple Extraction

RFBiCF: A Relation-First Bidirectional Cascade Framework for...

引用

Pattern Recognition and Machine Learning (PRML), IEEE International Conference on

作者： Deng Jinxin Qin Xizhong Yang Rong Lv Xiaoyi School of Information Science and Engineering Xinjiang University of China Urumqi China Xinjiang Key Laboratory of Signal Detection and Processing Urumqi China School of Information The Fifth Affiliated Hospital of Xinjiang Medical Urumqi China

Relational triple extraction from unstructured text, which is in form of (subject, relation, object), is crucial in information extraction and knowledge graph construction. However, most existing methods still suffer from some limitations, such as error propagation, redundancy of relation prediction, and poor generalization. To overcome these deficiencies, we propose a relation-first bidirectional cascade multi-task framework for relational triple extraction. This framework solves the relational redundancy problem by relation prediction, alleviates error propagation through a complementary bidirectional entity tagger, and uses multi-task learning to improve the generalization of the model. Concretely, we first use a relation detection component to predict potential relations in sentences and conduct the following entity pairs extraction based on specific relations. Then, we extract entity pairs from two paralleled directions. Meanwhile, entity recognition is used as an auxiliary task, improving generalization capability. Extensive experiments were conducted on the NYT and WebNLG datasets We evaluate our model on four public datasets to demonstrate the effectiveness of our proposed RFBiCF framework, even in the complex scenario of overlapping triples.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Research on Speech Synthesis Based on Prosodic Controllability

Research on Speech Synthesis Based on Prosodic Controllabili...

引用

Human-Computer Interaction, Optimization and Robotic Applications (HORA), International Congress on

作者： Wenyu Gao Askar Hamdulla Xipeng Yang Xinjiang Key Laboratory of Signal Detection and Processing School of Information Science and Engineering Xinjiang University of China Urumqi Xinjiang China Xinjiang Key Laboratory of Multilingual Information Technology School of Information Science and Engineering Xinjiang University of China Urumqi Xinjiang China Mobvoi AI Lab Suzhou China

In today's highly interactive human-computer world, speech synthesis is widely used in many scenarios, and the requirements for rhyme effects in speech synthesis technology are increasing, so rhyme-controllable models have become a research hotspot. Most current rhyme controllable models are achieved by using a separate neural network to generate reference features, but this approach requires the training of more complex neural network models and the availability of reference audio to achieve display rhyme control. This paper proposes a rhyme-controllable solution based on an end-to-end acoustic model to address the problem of models being unable to precisely control the tone at the word level. The proposed model includes a tone-controllable module, which obtains duration information through the MFA alignment tool and adjusts the tone of the words by using word-level pitch control values and duration information. The acoustic model in this paper is improved by introducing pitch control during the generation of acoustic features and generating more robust audio by combining it with the decoder. In addition, to adjust the overall tone of the audio, a fixed coefficient is multiplied to the pitch values of all frames. Furthermore, this paper also proposes a 48KHz ultra-high-definition audio model by increasing the spectral parameter dimensions and upsampling by a factor.

关键词：

来源：评论

学校读者我要写书评

暂无评论

LE-CAM++: A Lighter and More Efficient CAM++ for Speaker Verification

LE-CAM++: A Lighter and More Efficient CAM++ for Speaker Ver...

引用

International Symposium on Chinese Spoken Language processing

作者： Shuanghong Liu Zhida Song Zhihua Fang Liang He School of Computer Science and Technology Xinjiang University Urumqi Xinjiang Key Laboratory of Signal Detection and Processing Urumqi School of Intelligence Science and Technology Xinjiang University Urumqi Department of Electronic Engineering and Beijing National Research Center for Information Science and Technology Tsinghua University Beijing

ISBN: (数字)9798331516826

ISBN: (纸本)9798331516833

Due to its superior performance and fewer parameters, CAM++ has become the state-of-the-art model for speaker verification tasks. This model uses 2D convolutional blocks to extract front-end features, which are then fed into a densely connected time-delay neural network backbone to extract deep features. However, the simple stacking of 2D convolutions may lead to the generation of a significant amount of redundant features, which is detrimental to efficient feature extraction. Furthermore, although CAM++ already has a relatively small number of parameters, there is still room for further optimization. To address these issues, this paper first employs depthwise separable convolutions to replace the dilated convolutions in the back-end network of CAM++, making the model more lightweight. Next, we introduce spatial and channel reconstruction convolution (SCConv) in the ResBlock module of CAM++ to reduce redundant features and optimize the feature extraction process. Finally, after SCConv, we apply squeeze and excitation attention mechanism to model the interdependencies between channels and recalibrate each channel, further enhancing the model's representational capacity. We name the resulting model LE-CAM++. Our proposed model achieves an EER of 0.686 and a minDCF of 0.084 on the VoxCeleb1–O dataset. Compared to the baseline model CAM ++, the EER is reduced by 11%, and the minDCF is reduced by 28%. Additionally, the model parameters are reduced by 8%.

关键词： Attention mechanisms Stacking Neural networks Feature extraction Optimization

来源：评论

学校读者我要写书评

暂无评论

Simplified Skip-Connected UNet for Robust Speaker Verification Under Noisy Environments

Simplified Skip-Connected UNet for Robust Speaker Verificati...

引用

International Symposium on Chinese Spoken Language processing

作者： Zonghui Wang Zhihua Fang Zhida Song Liang He School of Computer Science and Technology Xinjiang University Urumqi Xinjiang Key Laboratory of Signal Detection and Processing Urumqi School of Intelligence Science and Technology Xinjiang University Urumqi Department of Electronic Engineering Beijing National Research Center for Information Science and Technology Tsinghua University Beijing

ISBN: (数字)9798331516826

ISBN: (纸本)9798331516833

In recent years, deep neural network based methods for speaker verification have made remarkable progress in clean environments. However, background noise significantly reduces the accuracy and reliability of speaker verification systems by masking or changing the voice characteristics of the speaker. In this paper, we propose a cascaded framework optimized with multiobjective loss to mitigate the interference of different levels and types of noise on the speaker verification task. The proposed architecture consists of two components: a speech enhancement module based on improved 2D-UNet, which reduces the structural limitations of directly using classical UNet for noise reduction, and a back-end speaker embedding extraction module. We carry our experiments on the VoxCeleb1 and VOiCES datasets, as well as in the presence of out-of-domain noise conditions. The evaluations have demonstrated this method shows great potential for speaker verification in noisy environments.

关键词： Training Semantics Noise reduction Interference Speech enhancement Complexity theory Noise measurement Reliability Optimization signal to noise ratio

来源：评论

学校读者我要写书评

暂无评论

A High Isolation Design Used Between Broadband Dual Polarization Transceiver Antennas

A High Isolation Design Used Between Broadband Dual Polariza...

引用

International Symposium on Antennas, Propagation and EM Theory (ISAPE)

作者： Yuke Guo Xiaosheng Zhang Luyu Zhao National Key Laboratory of Radar Detection and Sensing Xidian University Xi'an China Key Laboratory of Intelligent Computing and Signal Processing Ministry of Education Anhui University Hefei China

ISBN: (数字)9798350353129

ISBN: (纸本)9798350353136

We proposed a high isolation design between two broadband dual polarization transceiver antennas in this paper. It employs periodic metal isolation walls with equal spacing and height to strongly suppress the low-frequency band (LB) coupling. Covering the vertical and horizontal surfaces of the metal with flexible graphene high resistive metasurfaces that easily conform to the structure improves 40 dB in isolation between different polarizations at 6–18 GHz. This device has less impact on the radiation pattern of the antenna. This design not only enhances the isolation between different polarizations but also optimizes equipment heat dissipation and improves device performance.

关键词： Performance evaluation Heating systems Couplings Graphene Metals Metasurfaces Transceivers Broadband communication Broadband antennas Antenna radiation patterns

来源：评论

学校读者我要写书评

暂无评论

ZCS-CDiff: A Zero-Shot Code-Switching TTS System with Conformer-Based Diffusion Model

ZCS-CDiff: A Zero-Shot Code-Switching TTS System with Confor...

引用

International Conference on Acoustics, Speech, and signal processing (ICASSP)

作者： Ke Chen Zhihua Huang Liang He Yonghong Yan School of Computer Science and Technology Xinjiang University Urumqi China Xinjiang Key Laboratory of Signal Detection and Processing Urumqi China Department of Electronic Engineering Tsinghua University Beijing China University of Chinese Academy of Sciences Beijing China Key Laboratory of Speech Acoustics and Content Understanding Institute of Acoustics CAS Beijing China

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Code-Switching (CS) Text-To-Speech (TTS) models have gained attention due to the increasing prevalence of multilingual communication. However, existing models struggle to meet the growing demand for personalized CS TTS, particularly in Zero-Shot (ZS) scenarios where the model should generate speech for an unseen speaker from reference speech. In this paper, we propose ZCS-CDiff, a zero-shot code-switching TTS system with the Conformer-based diffusion model. We disentangle speech features and use the diffusion model to precisely model these disentangled attributes, resulting in high-quality ZS CS speech. Additionally, we introduce a Conformer-based WaveNet as the denoising network within the diffusion model to further improve the accuracy of modeling different attributes. We also designed a speaker-assist module to help the model better handle speaker information extracted from the reference speech, resulting in CS speech with higher speaker similarity. Experimental results and ablation studies demonstrate that the ZS CS speech generated by ZCS-CDiff has good speech naturalness, intelligibility, and speaker similarity while confirming the effectiveness of our design choices. Audio samples are available 1 .

关键词： Training Adaptation models Accuracy Speech coding Noise reduction Speech enhancement signal processing Diffusion models Probabilistic logic Text to speech

来源：评论

学校读者我要写书评

暂无评论

A Singing Melody Extraction Network Via Self-Distillation and Multi-Level Supervision

A Singing Melody Extraction Network Via Self-Distillation an...

引用

International Conference on Acoustics, Speech, and signal processing (ICASSP)

作者： Ying Hu Jiabo Jing Fan Li Lijun He Li Lin Wenzhong Yang Key Laboratory of Signal Detection and Processing Urumqi China School of Computer Science and Technology Xinjiang University Urumqi China School of Information and Communications Engineering Xi’an Jiaotong University Xi’an China Xinjiang Institute of Electronics Research Shares Co. Ltd Urumgi China

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

Extracting singing melody from polyphonic music is an important topic in the field of music information retrieval. In this paper, we propose a singing melody extraction network consisting of five stacked multi-scale feature time-frequency aggregation (MF-TFA) modules. In the same network, deeper layers generally contain more contextual information than shallower layers. To help the shallower layers enhance the ability of task-relevant feature extraction, we propose a self-distillation and multi-level supervision (SD-MS) method, which leverages the feature distillation from the deepest layer to the shallower one and multi-level supervision to guide network training. Visualization analysis shows that by introducing SD-MS, the same-level layer in the network can obtain a clearer representation of fundamental frequency components, while the shallower layers can even learn more task-relevant semantic information. Ablation study results indicate that SD-MS applies to existing melody extraction models and can consistently improve performance. Experimental results show that our proposed method, MF-TFA with SD-MS, outperforms six compared state-of-the-art methods, achieving overall accuracy (OA) scores of 87.1%, 89.9%, and 76.6% on the ADC 2004, MIREX 05, and MEDLEY DB datasets, respectively. The main code will be available at https://***/SmoothJing/MF-TFA_SD-MS.

关键词： Training Visualization Time-frequency analysis Semantics signal processing Feature extraction Multiple signal classification Data mining Speech processing Music information retrieval

来源：评论

学校读者我要写书评

暂无评论

Feature-Consistent Cyclic Domain Adversarial Network for Unsupervised Transfer Diagnosis of the Autoimmune Disease Igg4-Rd by Raman Spectroscopy

SSRN

引用

SSRN 2025年

作者： Zhang, Ziyang Chen, Cheng Qin, Wentao Wu, Xue Liu, Yang Wu, Lijun Lv, Xiaoyi Chang, Chenjie Chen, Xinya Chen, Zhiyuan Xu, Qianxi Li, Zhuoyu Ma, Mengsi Chen, Chen Xinjiang University College of Software Urumqi830046 China Xinjiang Cloud Computing Application Laboratory Xinjiang University Karamay834099 China School of Computer Science and Technology Xinjiang University Urumqi830046 China Key Laboratory of signal detection and processing Xinjiang University Urumqi830046 China Department of Rheumatology＆Immunology People's Hospital of Xinjiang Uygur Autonomous Region Xinjiang Urumqi China Xinjiang Clinical Research Center for Rheumatoid arthritis Xinjiang Urumqi China

Abstracts IgG4-related disease (IgG4-RD) is a multi-organ immune disorder characterized by systemic involvement, diverse pathogenesis, and rarity, which complicates its diagnosis. Traditional medical diagnostic models, predominantly based on supervised learning in machine learning or deep learning frameworks, struggle to address the scarcity of labeled data due to the diagnostic challenges of IgG4-RD. To overcome this limitation, this study proposes a novel framework based on Feature-Consistent Cyclic Domain Adversarial Networks (FCCDAN). The framework first trains a classification model on the source domain for diagnosing source domain diseases under supervised conditions. Subsequently, leveraging the source domain model, the encoder is guided to extract semantic features, while feature consistency loss is integrated into both the generator and discriminator to align the semantic features of the target domain with those of the source domain. Finally, the aligned features are processed through a pre-trained linear mapping to achieve unsupervised classification and diagnosis in the target domain. The proposed model utilizes common autoimmune diseases as a knowledge source, enabling unsupervised transfer diagnosis for IgG4-RD, and demonstrates generalizability through experiments on other disease datasets, achieving an average diagnostic accuracy of 90.62%. This represents improvements of 4.14%, 9.82%, and 10.51% over baseline models, respectively. FCCDAN expands the applicability of Raman spectroscopy in disease diagnosis, maintaining high performance even in label-scarce environments. The overall research workflow also establishes a robust paradigm for the application and advancement of domain adaptation techniques in this context. © 2025, The Authors. All rights reserved.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Exploring Turkish Speech Recognition via Hybrid CTC/Attention Architecture and Multi-feature Fusion Network

Exploring Turkish Speech Recognition via Hybrid CTC/Attentio...

引用

Frontiers of Electronics, Information and Computation Technologies (ICFEICT), International Conference on

作者： Nurmement Yolwas Zeyu Ren Huiru Wang Wushour Slamu Xinjiang Multilingual Information Technology Laboratory College of Information Science and Engineering Xinjiang University Urumqi China Xinjiang Key Laboratory of Signal Detection College of Information Science and Engineering Xinjiang University Urumqi China

In recent years, there has been rapid development in deep learning-based End-to-End speech recognition technology. However, the performance of Turkish speech recognition systems has been hindered by the lack of Turkish speech data. To address this, this paper focuses on several speech recognition tuning techniques. Experimental results demonstrate that the best performance is achieved when utilizing a combination of speed perturbation and noise addition for data augmentation, along with the beam search width set to 16. Furthermore, to maximize the utilization of effective feature information and enhance feature extraction accuracy, a novel feature extractor called LSPC is proposed. By combining LSPC with the LiGRU network, a shared encoder structure is formed, and model compression is achieved. The results indicate that when using only Fbank features, the performance of LSPC surpasses that of MSPC and VGGnet, resulting in a respective improvement of 1.01% and 2.53% in the word error rate (WER). Finally, building upon the aforementioned advancements, a new multifeature fusion network is proposed as the primary encoder structure. The results demonstrate that the WER of the proposed feature fusion network, based on LSPC, further improves by 0.82% and 1.94% when compared to single-feature extraction using LSPC, specifically utilizing Fbank features and spectrogram features. As a result, our model achieves comparable performance to advanced End-to-End models.

关键词：

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：