检索结果-内蒙古大学图书馆

Topic-enhanced emotional conversation generation with attention mechanism

KNOWLEDGE-BASED SYSTEMS 2019年第Jan.1期163卷 429-437页

作者： Peng, Yehong Fang, Yizhen Xie, Zhiwen Zhou, Guangyou Cent China Normal Univ Sch Comp Wuhan 430079 Hubei Peoples R China

Emotional conversation generation has elicited a wide interest in both academia and industry. However, existing emotional neural conversation systems tend to ignore the necessity to combine topic and emotion in generating responses, possibly leading to a decline in the quality of responses. This paper proposes a topic-enhanced emotional conversation generation model that incorporates emotional factors and topic information into the conversation system, by using two mechanisms. First, we use a Twitter latent Dirichlet allocation (LDA) model to obtain topic words of the input sequences as extra prior information, ensuring the consistency of content between posts and responses for emotional conversation generation. Second, the system uses a dynamic emotional attention mechanism to adaptively acquire content-related and affective information of the input texts and extra topics. The advantage of this study lies in the fact that the presented model can generate abundant emotional responses, with the contents being related and diverse. To demonstrate the effectiveness of our method, we conduct extensive experiments on large-scale Weibo post-response pairs. Experimental results show that our method achieves good performance, even outperforming some existing models. (C) 2018 Elsevier B.V. All rights reserved.

关键词： Emotional conversation Topic model sequence-to-sequence Attention mechanism

来源：评论

学校读者我要写书评

暂无评论

An encoder-decoder based grapheme-to-phoneme converter for Bangla speech synthesis

引用

ACOUSTICAL SCIENCE AND TECHNOLOGY 2019年第6期40卷 374-381页

作者： Ahmad, Arif Selim, Mohammad Reza Iqbal, Muhammed Zafar Rahman, Mohammad Shahidur Shahjalal Univ Sci & Technol Dept Comp Sci & Engn Sylhet 3114 Bangladesh

This paper proposes an encoder-decoder based sequence-to-sequence model for Grapheme-to-Phoneme (G2P) conversion in Bangla (Exonym: Bengali). G2P models are key components in speech recognition and speech synthesis systems as they describe how words are pronounced. Traditional, rule-based models do not perform well in unseen contexts. We propose to adopt a neural machine translation (NMT) model to solve the G2P problem. We used gated recurrent units (GRU) recurrent neural network (RNN) to build our model. In contrast to joint-sequence based G2P models, our encoder-decoder based model has the flexibility of not requiring explicit grapheme-to-phoneme alignment which are not straight forward to perform. We trained our model on a pronunciation dictionary of (approximately) 135,000 entries and obtained a word error rate (WER) of 12.49% which is a significant improvement from the existing rule-based and machine-learning based Bangla G2P models.

关键词： Encoder-decoder sequence-to-sequence GRU-RNN NMT

来源：评论

学校读者我要写书评

暂无评论

End-to-End Training for Compound Expression Recognition

引用

SENSORS 2020年第17期20卷 4727.-4727.页

作者： Li, Hongfei Li, Qing Chinese Acad Sci Inst Microelect Beijing 100029 Peoples R China Univ Chinese Acad Sci Beijing 100049 Peoples R China

For a long time, expressions have been something that human beings are proud of. That is an essential difference between us and machines. With the development of computers, we are more eager to develop communication between humans and machines, especially communication with emotions. The emotional growth of computers is similar to the growth process of each of us, starting with a natural, intimate, and vivid interaction by observing and discerning emotions. Since the basic emotions, angry, disgusted, fearful, happy, neutral, sad and surprised are put forward, there are many researches based on basic emotions at present, but few on compound emotions. However, in real life, people's emotions are complex. Single expressions cannot fully and accurately show people's inner emotional changes, thus, exploration of compound expression recognition is very essential to daily life. In this paper, we recommend a scheme of combining spatial and frequency domain transform to implement end-to-end joint training based on model ensembling between models for appearance and geometric representations learning for the recognition of compound expressions in the wild. We are mainly devoted to digging the appearance and geometric information based on deep learning models. For appearance feature acquisition, we adopt the idea of transfer learning, introducing the ResNet50 model pretrained on VGGFace2 for face recognition to implement the fine-tuning process. Here, we try and compare two minds, one is that we utilize two static expression databases FER2013 and RAF Basic for basic emotion recognition to fine tune, the other is that we fine tune the model on the input three channels composed of images generated by DWT2 and WAVEDEC2 wavelet transforms based on rbio3.1 and sym1 wavelet bases respectively. For geometric feature acquisition, we firstly introduce a densesift operator to extract facial key points and their histogram descriptions. After that, we introduce deep SAE with a softmax func

关键词： joint training end-to-end deep SAE compound expression appearance feature geometric feature model ensembling frequency domain transform stacked LSTM sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

Concurrent phonetic transcription, lexical stress assignment and syllabification with deep neural networks

引用

Procedia Computer Science 2020年 176卷 108-117页

作者： Beáta Lőrincz Technical University Communications Department 400027 Cluj-Napoca Romania Babeş-Bolyai University Faculty of Mathematics and Computer Science 400084 Cluj-Napoca Romania

This paper evaluates four different sequence-to-sequence deep neural network architectures aimed to jointly solve the tasks of: phonetic transcription, lexical stress assignment and syllabification. These text processing tasks are considered essential components for high quality text-to-speech or automatic speech recognition systems, with the phonetic transcription being the most frequently used in these types of applications. Although each of the tasks has been individually and extensively analyzed in the scientific literature, there are few studies which target a concurrent solution for them. In general, the lexical stress assignment and syllabification are used as augmenting input features to the phonetic transcription model and not considered as target features. The proposed network architectures include recurrent, convolution and attention neural layers and were evaluated on hand-checked English and Romanian datasets. The accuracy of the models was evaluated in terms of accuracy for the concurrent prediction of all three tasks, as well as by discarding the syllabification or lexical stress predictions. The best results were obtained with a combination of convolution and attention layers, where the accuracy of the joint prediction for the three tasks was of 58.96% for English and 86.64% for Romanian. The same model for English obtains an accuracy of 59.70% when syllables are discarded and 64% when the prediction of lexical stress is ignored. With the same best performing model for Romanian an accuracy of 88.83% without syllables and 93.84% without lexical stress is obtained.

关键词： phonetic transcription syllabification lexical stress assignment neural networks sequence-to-sequence English Romanian

来源：评论

学校读者我要写书评

暂无评论

A SPELLING CORRECTION MODEL FOR END-TO-END SPEECH RECOGNITION 44

A SPELLING CORRECTION MODEL FOR END-TO-END SPEECH RECOGNITIO...

引用

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

作者： Guo, Jinxi Sainath, Tara N. Weiss, Ron J. Univ Calif Los Angeles Los Angeles CA 90095 USA Google Inc Mountain View CA USA

ISBN: (纸本)9781479981311

Attention-based sequence-to-sequence models for speech recognition jointly train an acoustic model, language model (LM), and alignment mechanism using a single neural network and require only parallel audio-text pairs. Thus, the language model component of the end-to-end model is only trained on transcribed audio-text pairs, which leads to performance degradation especially on rare words. While there have been a variety of work that look at incorporating an external LM trained on text-only data into the end-to-end framework, none of them have taken into account the characteristic error distribution made by the model. In this paper, we propose a novel approach to utilizing text-only data, by training a spelling correction (SC) model to explicitly correct those errors. On the LibriSpeech dataset, we demonstrate that the proposed model results in an 18.6% relative improvement in WER over the baseline model when directly correcting top ASR hypothesis, and a 29.0% relative improvement when further rescoring an expanded n-best list using an external LM.

关键词： speech recognition sequence-to-sequence attention models spelling correction language model

来源：评论

学校读者我要写书评

暂无评论

Fine-Grained Robust Prosody Transfer for Single-Speaker Neural Text-To-Speech 20

Fine-Grained Robust Prosody Transfer for Single-Speaker Neur...

引用

Interspeech Conference

作者： Klimkov, Viacheslav Ronanki, Srikanth Rohnke, Jonas Drugman, Thomas Amazon Res Cambridge MA 02139 USA

We present a neural text-to-speech system for fine-grained prosody transfer from one speaker to another. Conventional approaches for end-to-end prosody transfer typically use either fixed-dimensional or variable-length prosody embedding via a secondary attention to encode the reference signal. However, when trained on a single-speaker dataset, the conventional prosody transfer systems are not robust enough to speaker variability, especially in the case of a reference signal coming from an unseen speaker. Therefore, we propose decoupling of the reference signal alignment from the overall system. For this purpose, we pre-compute phoneme-level time stamps and use them to aggregate prosodic features per phoneme, injecting them into a sequence-to-sequence text-to-speech system. We incorporate a variational auto-encoder to further enhance the latent representation of prosody embeddings. We show that our proposed approach is significantly more stable and achieves reliable prosody transplantation from an unseen speaker. We also propose a solution to the use case in which the transcription of the reference signal is absent. We evaluate all our proposed methods using both objective and subjective listening tests.

关键词： Neural text-to-speech sequence-to-sequence prosody transfer

来源：评论

学校读者我要写书评

暂无评论

Enhancing Conversational Dialogue Models with Grounded Knowledge 19

Enhancing Conversational Dialogue Models with Grounded Knowl...

引用

28th ACM International Conference on Information and Knowledge Management (CIKM)

作者： Zheng, Wen Zhou, Ke Univ Nottingham Nottingham England Nokia Bell Labs Cambridge England

ISBN: (纸本)9781450369763

Leveraging external knowledge to enhance conversational models has become a popular research area in recent years. Compared to vanilla generative models, the knowledge-grounded models may produce more informative and engaging responses. Although various approaches have been proposed in the past, how to effectively incorporate knowledge remains an open research question. It is unclear how much external knowledge should be retrieved and what is the optimal way to enhance the conversational model, trading off between relevant information and noise. Therefore, in this paper, we aim to bridge the gap by first extensively evaluating various types of state-of-the-art knowledge-grounded conversational models, including recurrent neural network based, memory networks based, and Transformer based models. We demonstrate empirically that those conversational models can only be enhanced with the right amount of external knowledge. To effectively leverage information originated from external knowledge, we propose a novel Transformer with Expanded Decoder (Transformer-ED or TED for short), which can automatically tune the weights for different sources of evidence when generating responses. Our experiments show that our proposed model outperforms state-of-the-art models in terms of both quality and diversity.

关键词： sequence-to-sequence Copy-Mechanism Memory Network Multi-Task Learning Transformer Transformer-ED TED Generative Model Knowledge-Grounded

来源：评论

学校读者我要写书评

暂无评论

Improving a Syntactic Graph Convolution Network for Sentence Compression 18th

Improving a Syntactic Graph Convolution Network for Sentence...

引用

18th China National Conference on Computational Linguistics (CCL)

作者： Wang, Yifan Chen, Guang Beijing Univ Posts & Telecommun Beijing Peoples R China

ISBN: (纸本)9783030323813;9783030323806

Sentence compression is a task of compressing sentences containing redundant information into short semantic expressions, simplifying the text structure and retaining important meanings and information. Neural network-based models are limited by the size of the window and do not perform well when using long-distance dependent information. To solve this problem, we introduce a version of the graph convolutional network (GCNs) to utilize the syntactic dependency relations, and explore a new way to combine GCNs with the sequence-to-sequence model (Seq2Seq) to complete the task. The model combines the advantages of both and achieves complementary effects. In addition, in order to reduce the error propagation of the parse tree, we dynamically adjust the dependency arc to optimize the construction process of GCNs. Experiments show that the model combined with the graph convolution network is better than the original model, and the performance in the Google sentence compression dataset has been effectively improved.

关键词： Sentence compression Graph convolution network sequence-to-sequence

来源：评论

学校读者我要写书评

暂无评论

AAT: An Efficient Model for Synthesizing Long sequences on a Small Dataset 13th

AAT: An Efficient Model for Synthesizing Long Sequences on a...

引用

13th Multi-Disciplinary International Conference on Multi-Disciplinary Trends in Artificial Intelligence (MIWAI)

作者： Quan Anh Minh Nguyen Quang Trinh Le Huu Quoc Van Duc Dung Nguyen HO CHI MINH City Univ Technol Ho Chi Minh City Vietnam

ISBN: (纸本)9783030337094;9783030337087

This work represents an alternative model for speech synthesis, which addresses some major disadvantages of current end-to-end models. Current state-of-the-art models still have some troubles while dealing with long sentences and the size of the dataset. Our proposed Adaptive Alignment Tacotron (AAT) model, however, has achieved impressive results for a very small dataset of the Vietnamese language. By leveraging the nature of diagonal alignment between phoneme and acoustic sequences, we address the issue with long sequences and the proposed model can also be trained efficiently on the small dataset (5 h of recording). The proposed model consists of the following components: a stacked convolutional encoder, a local diagonal attention module, a decoder with schedule teacher forcing to produce a coarse mel-spectrogram prediction, and a converter to transform the melspectrogram to linear-spectrogram. Experimental results show that the proposed model achieves faster convergence speed and higher stability than the baseline model and open a feasible approach for speech synthesis on languages with small dataset.

关键词： Text-to-speech Attention Encoder-decoder sequence-to-sequence Tacotron Vietnamese speech synthesis

来源：评论

学校读者我要写书评

暂无评论

Keyphrase Generation with Word Attention 26th

Keyphrase Generation with Word Attention

引用

26th International Conference on Neural Information Processing (ICONIP) of the Asia-Pacific-Neural-Network-Society (APNNS)

作者： Huang, Hai Huang, Tianshuo Ma, Longxuan Zhang, Lei Beijing Univ Posts & Telecommun Beijing Peoples R China

ISBN: (纸本)9783030367183;9783030367176

Keyphrase generation aims to generate several words that can simply summarize the semantics of the article, which is the basis of many natural language processing tasks. Although most previous approaches have achieved good results, they neglect the independent word information in the source text. Previous models use attention mechanism to calculate the relationship between the encoder RNN hidden states and the target side. However hidden state ht is the summarization of the first t words as a subsequence in the source sentence. We, in this paper, propose a novel sequence-to-sequence model called WordRNN, which can capture word level representations in the source text. Our model can enrich the expression of the source text by directly promoting the pure word level information. Moreover, we use fuse gate and simply concat operation to combine the subsequence level and word level contextual information. Experiment results demonstrate that our approach achieves higher performance than the state-of-the-art methods.

关键词： Keyphrase generation sequence-to-sequence Attention

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：