检索结果-内蒙古大学图书馆

44th IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

作者： Huang, Kun-Yi Wu, Chung-Hsien Hong, Qian-Bei Su, Ming-Hsiang Chen, Yi-Hsuan Natl Cheng Kung Univ Dept Comp Sci & Informat Engn Tainan Taiwan

ISBN: (纸本)9781479981311

Speech emotion recognition is becoming increasingly important for many applications. In real-life communication, non-verbal sounds within an utterance also play an important role for people to recognize emotion. In current studies, only few emotion recognition systems considered nonverbal sounds, such as laughter, cries or other emotion interjection, which naturally exists in our daily conversation. In this work, both verbal and nonverbal sounds within an utterance were thus considered for emotion recognition of real-life conversations. Firstly, an SVM-based verbal/nonverbal sound detector was developed. A Prosodic Phrase ( PPh) auto-tagger was further employed to extract the verbal/ nonverbal segments. For each segment, the emotion and sound features were respectively extracted based on convolutional neural networks ( CNNs) and then concatenated to form a CNN-based generic feature vector. Finally, a sequence of CNN-based feature vectors for an entire dialog turn was fed to an attentive long short-term memory ( LSTM)-based sequence-to-sequence model to output an emotional sequence as recognition result. Experimental results on the recognition of seven emotional states in the NNIME ( The NTHU-NTUA Chinese interactive multimodal emotion corpus) showed that the proposed method achieved a detection accuracy of 52.00% outperforming the traditional methods.

关键词： Speech emotion recognition prosodic Phrase nonverbal segment convolutional neural network long-short term memory sequence-to-sequence model

来源：评论

学校读者我要写书评

暂无评论

Sentence Simplification Based on Multi-Stage Encoder model

引用

IEEE ACCESS 2019年 7卷 174248-174256页

作者： Zhang, Lemin Deng, Huifang South China Univ Technol Sch Comp Sci & Engn Guangzhou 510006 Peoples R China

Sentence simplification aims to simplify a complex sentences while retaining its main idea. It is one of the most important tasks in natural language processing. Recent works addressed the task with sequence-to-sequence(Seq2seq) model. However, these conventional Seq2seq models usually based on a single-stage encoder, which only read the source complex sentence once, as a result, it was hard to extract the representational features of the source sentence precisely. To resolve the problem, we proposed a multi-stage encoder based Seq2seq model for sentence simplification. Specificly, there were three stages in the encoder of proposed model, namely N-gram reading stage, glance-over stage and final encoding stage. The N-gram reading stage catched N-gram feature embedding for other stage and the glance-over stage extracted local and global information about the source sentence. The final encoding stage took advantage of the information extracted by the former two stage to encode source sentence better. Then, it introduced a novel attention connection method which could help the decoder to make full use of the information of encoder. Experiments on three public datasets demonstrated the proposed model that outperforms state-ofthe-art baseline simplification systems.

关键词： Sentence simplification natural language processing sequence-to-sequence model

来源：评论

学校读者我要写书评

暂无评论

Nonintrusive Load Monitoring Based on Deep Learning 1

引用

6th ECML PKDD International Workshop on Data Analytics for Renewable Energy Integration (DARE)

作者： Wang, Ke Zhong, Haiwang Yu, Nanpeng Xia, Qing Tsinghua Univ Dept Elect Engn Beijing 100084 Peoples R China Tsinghua Univ State Key Lab Control & Simulat Power Syst & Gene Beijing 100084 Peoples R China Univ Calif Riverside Riverside CA 92521 USA

ISBN: (数字)9783030043032

ISBN: (纸本)9783030043032;9783030043025

This paper presents a novel nonintrusive load monitoring method based on deep learning. Unlike the existing work based on convolutional neural network and recurrent neural network with fully connected layers, this paper develops a deep neural network based on sequence-to-sequence model and attention mechanism to perform nonintrusive load monitoring. The overall framework can be divided into three layers. In the first layer, the input active power time sequence is embedded into a group of high dimensional vectors. In the second layer, the vectors are encoded by a bi-directional LSTM layer, and the N encoded vectors are added up to form a dynamic context vector according to its weights calculated by the attention mechanism. In the third layer, an LSTM-based decoder utilizes the dynamic context vector to calculate the disaggregated power consumption at every time step. The proposed method is trained and tested on REFITPowerData dataset. The results show that compared to the state-of-the-art methods, the proposed method significantly increases the accuracy of the estimation for the disaggregated power value and decreases the misjudge rate by 10% to 20%.

关键词： Nonintrusive load monitoring Deep learning sequence-to-sequence model Attention mechanism

来源：评论

学校读者我要写书评

暂无评论

FORWARD ATTENTION IN sequence-TO-sequence ACOUSTIC modelING FOR SPEECH SYNTHESIS

FORWARD ATTENTION IN SEQUENCE-TO-SEQUENCE ACOUSTIC MODELING ...

引用

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

作者： Zhang, Jing-Xuan Ling, Zhen-Hua Dai, Li-Rong Univ Sci & Technol China Natl Engn Lab Speech & Language Informat Proc Hefei Anhui Peoples R China

ISBN: (纸本)9781538646588

This paper proposes a forward attention method for the sequence-to-sequence acoustic modeling of speech synthesis. This method is motivated by the nature of the monotonic alignment from phone sequences to acoustic sequences. Only the alignment paths that satisfy the monotonic condition are taken into consideration at each decoder timestep. The modified attention probabilities at each timestep are computed recursively using a forward algorithm. A transition agent for forward attention is further proposed, which helps the attention mechanism to make decisions whether to move forward or stay at each decoder timestep. Experimental results show that the proposed forward attention method achieves faster convergence speed and higher stability than the baseline attention method. Besides, the method of forward attention with transition agent can also help improve the naturalness of synthetic speech and control the speed of synthetic speech effectively.

关键词： sequence-to-sequence model encoder-decoder attention speech synthesis

来源：评论

学校读者我要写书评

暂无评论

END-TO-END MULTIMODAL SPEECH RECOGNITION

END-TO-END MULTIMODAL SPEECH RECOGNITION

引用

IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

作者： Palaskar, Shruti Sanabria, Ramon Metze, Florian Carnegie Mellon Univ Pittsburgh PA 15213 USA

ISBN: (纸本)9781538646588

Transcription or sub-titling of open-domain videos is still a challenging domain for Automatic Speech Recognition (ASR) due to the data's challenging acoustics, variable signal processing and the essentially unrestricted domain of the data. In previous work, we have shown that the visual channel - specifically object and scene features - can help to adapt the acoustic model (AM) and language model (LM) of a recognizer, and we are now expanding this work to end-to-end approaches. In the case of a Connectionist Temporal Classification (CTC)-based approach, we retain the separation of AM and LM, while for a sequence-to-sequence (S2S) approach, both information sources are adapted together, in a single model. This paper also analyzes the behavior of CTC and S2S models on noisy video data (How-To corpus), and compares it to results on the clean Wall Street Journal (WSJ) corpus, providing insight into the robustness of both approaches.

关键词： Audiovisual Speech Recognition Connectionist Temporal Classification sequence-to-sequence model Adaptation

来源：评论

学校读者我要写书评

暂无评论

Knowledge-based Questions Generation with Seq2Seq Learning 6

Knowledge-based Questions Generation with Seq2Seq Learning

引用

IEEE International Conference on Progress in Informatics and Computing (PIC)

作者： Tang, Xiangru Gao, Hanning Gao, Junjie Peking Univ Inst Comp Sci & Technol Beijing Peoples R China Cent China Normal Univ Sch Comp Wuhan Hubei Peoples R China

ISBN: (纸本)9781538676721

Building human-computer interaction engines is a fundamental challenge in Artificial Intelligence, with the goal of building a system which can automatically talk to human in natural language. Therefore, the ability to ask thought-provoking questions and contain valuable information in a conversational system is obviously quite crucial and challenging. In this paper, we build a knowledge-based question generation(KBQG) system which can generate natural language questions based on the given knowledge triples to demonstrate how large-scale knowledge base can facilitate language understanding and generation. We propose to apply the neural SEQ2SEQ learning model to generate meaningful questions from the given triples. The encoder reads in triples and produces an answer-aware input representation, which is fed to the decoder to generate an answer-focused question. To he able to express semantics completely and guarantee semantic integrity, we develop a flexible copy-attention mechanism that automatically replicates words from triples when needed. Both automatic and human evaluation results show that our model achieves outstanding performance. Furthermore, we establish a new evaluation metrics called GLAVE(Generated Language's automatic vector-pooling evaluation) to measure the similarity by pooling of word embeddings instead of word overlapping information, on account of the diversity and uncertainty of utterances. There are also a series of experiments which can demonstrate the effectiveness of our model is much stronger than state-of-the-art baselines.

关键词： language generation sequence-to-sequence model attention mechanism copynet Chinese knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Reconstructing Missing Data in Event Logs using Deep Learning Methods

Reconstructing Missing Data in Event Logs using Deep Learnin...

引用

作者： LINDA HAMP CHALMERS UNIVERSITY OF TECHNOLOGY

学位级别：硕士

Advances in computational power and deep learning methods enable many indus- tries to use their data to gain more meaningful insights than previously possible. The data is commonly found in event logs from software systems where events are delivered via network connections. Consequently, the deliveries could be adversely affected by network congestion or weak signals. Unless care is taken to resend data that is not successfully delivered, events can be lost. In this project, the aim is to reconstruct missing events using recurrent neural net- works. The models evaluated are a bidirectional recurrent neural network (BRNN) and a sequence-to-sequence model. Deep learning methods have previously been used to reconstruct missing data in sequences, but the locations of the gaps have then been declared in the input and are thus known in advance by the models. In this project, deep learning models have been trained to reconstruct incomplete se- quences where the locations of missing events are not given in advance. This makes them more useful for real applications. In our proposed method, the models are trained using supervised learning with complete sequences as target output. The input consists of both incomplete and complete sequences, where the incomplete sequences are constructed by removing events from complete ones. In this project, the dataset comprises event data pro- vided by the company where this thesis was conducted, and contains events related to the navigation of their application. Our findings are that the sequence-to-sequence model succeeds in reconstructing 75.42% of all missing events with a 95% confidence interval of [74.81%, 76.02%], whereas the BRNN is only able to reconstruct 64.99% of all missing events, with a 95% confidence interval of [64.16%, 65.83%]. From this we conclude that both models can be successfully used for the task of event reconstruction, with the sequence-to- sequence model being the best candidate.

关键词： deep learning bidirectional recurrent neural networks sequence-to-sequence model event logs reconstruction

来源：评论

学校读者我要写书评

暂无评论

FORWARD ATTENTION IN sequence-TO-sequence ACOUSTIC modelING FOR SPEECH SYNTHESIS

FORWARD ATTENTION IN SEQUENCE-TO-SEQUENCE ACOUSTIC MODELING ...

引用

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： Jing-Xuan Zhang Zhen-Hua Ling Li-Rong Dai National Engineering Laboratory for Speech and Language Information Processing University of Science and Technology of China Hefei P.R.China

ISBN: (纸本)9781538646595

关键词： sequence-to-sequence model encoder-decoder attention speech synthesis Speech synthesis acoustic modeling decoders Telephone modeling

来源：评论

学校读者我要写书评

暂无评论

END-TO-END MULTIMODAL SPEECH RECOGNITION

END-TO-END MULTIMODAL SPEECH RECOGNITION

引用

IEEE International Conference on Acoustics, Speech and Signal Processing

作者： Shruti Palaskar Ramon Sanabria Florian Metze Carnegie Mellon University Pittsburgh PA U.S.A.

ISBN: (纸本)9781538646595

关键词： Audiovisual Speech Recognition Connectionist Temporal Classification sequence-to-sequence model Adaptation Speech recognition recognizer Acoustic models Information source signal processing Open domains Genetic Transcription modelling languages

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：