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： Jia, Ye Johnson, Melvin Macherey, Wolfgang Weiss, Ron J. Cao, Yuan Chiu, Chung-Cheng Ari, Naveen Laurenzo, Stella Wu, Yonghui Google Res Mountain View CA 94043 USA

ISBN: (纸本)9781479981311

End-to-end Speech Translation (ST) models have many potential advantages when compared to the cascade of Automatic Speech Recognition (ASR) and text Machine Translation (MT) models, including lowered inference latency and the avoidance of error compounding. However, the quality of end-to-end ST is often limited by a paucity of training data, since it is difficult to collect large parallel corpora of speech and translated transcript pairs. Previous studies have proposed the use of pre-trained components and multi-task learning in order to benefit from weakly supervised training data, such as speech-to-transcript or text-to-foreign-text pairs. In this paper, we demonstrate that using pre-trained MT or text-to-speech (TTS) synthesis models to convert weakly supervised data into speech-to-translation pairs for ST training can be more effective than multi-task learning. Furthermore, we demonstrate that a high quality end-to-end ST model can be trained using only weakly supervised datasets, and that synthetic data sourced from unlabeled monolingual text or speech can be used to improve performance. Finally, we discuss methods for avoiding overfitting to synthetic speech with a quantitative ablation study.

关键词： Speech translation sequence-to-sequence model weakly supervised learning synthetic training data

来源：评论

学校读者我要写书评

暂无评论

A Context based Coverage model for Abstractive Document Summarization 10

A Context based Coverage Model for Abstractive Document Summ...

引用

10th International Conference on Information and Communication Technology Convergence (ICTC) - ICT Convergence Leading the Autonomous Future

作者： Kim, Heechan Lee, Soowon Soongsil Univ Dept Software Convergence Seoul South Korea Soongsil Univ Sch Software Seoul South Korea

ISBN: (纸本)9781728108933

Automatic abstractive summarization is one of natural language processing fields that generates a sequence of words representing important information of the input document. The sequence-to-sequence model, which is widely used in abstractive summarization, has a repetition problem in which the same sub-pattern is repeatedly generated in decode phases. To solve this problem, various coverage models have been proposed in machine translation. In automatic summarization, unlike machine translation, the lengths of input and summary documents are very different. Because the summary document is a compressed form of important meaning of the input document. Due to the nature of automatic summarization, it is difficult to apply a word position-based coverage model in machine translation directly. For automatic summarization, we propose a context based coverage model to consider the coverage based on the compressed meaning of the input document. The context based coverage is defined as the accumulated weighted average of the encoded word meaning by the attention scores. This considers the meaning of words rather than the position of words in the input document. In the experiment with CNN/DailyMail dataset, the proposed model shows better performance than the previous researches.

关键词： Context Coverage sequence-to-sequence model Abstractive Document Summarization Natural Language Processing

来源：评论

学校读者我要写书评

暂无评论

Parrotron: An End-to-End Speech-to-Speech Conversion model and its Applications to Hearing-Impaired Speech and Speech Separation 20

Parrotron: An End-to-End Speech-to-Speech Conversion Model a...

引用

Interspeech Conference

作者： Biadsy, Fadi Weiss, Ron J. Moreno, Pedro J. Kanvesky, Dimitri Jia, Ye Google Mountain View CA 94043 USA

We describe Parrotron, an end-to-end-trained speech-to-speech conversion model that maps an input spectrogram directly to another spectrogram, without utilizing any intermediate discrete representation. The network is composed of an encoder, spectrogram and phoneme decoders, followed by a vocoder to synthesize a time-domain waveform. We demonstrate that this model can be trained to normalize speech from any speaker regardless of accent, prosody, and background noise, into the voice of a single canonical target speaker with a fixed accent and consistent articulation and prosody. We further show that this normalization model can be adapted to normalize highly atypical speech from a deaf speaker, resulting in significant improvements in intelligibility and naturalness, measured via a speech recognizer and listening tests. Finally, demonstrating the utility of this model on other speech tasks, we show that the same model architecture can be trained to perform a speech separation task.

关键词： speech normalization voice conversion atypical speech speech synthesis sequence-to-sequence model

来源：评论

学校读者我要写书评

暂无评论

Exaggerated Portrait Caricatures Generation Based On Seq2Seq 9

Exaggerated Portrait Caricatures Generation Based On Seq2Seq

引用

9th International Conference on Information Science and Technology (ICIST)

作者： Xu, Kun Tang, Chenwei Lv, Jiancheng He, Zhenan Sichuan Univ Coll Comp Sci Chengdu Peoples R China

ISBN: (纸本)9781728121062

Exaggerated portrait caricatures can represent peoples explicit facial characteristics. Recently, there have been increasing interests in exaggerated portrait caricatures generation from the real face images. However, existing methods randomly exaggerate facial features and the effects are not diversified. In this paper, we achieve the extraction of facial structure features and explicit geometry features using sequence-to-sequence model. The extracted facial structure features can be well applied to caricatures generation. In order to avoid destroying the recognition of the characters from exaggerating the input face, we also incorporate the matching threshold loss as a penalty term to constrain the exaggerated expression of facial features. Experiment results on the collected dataset indicate the proposed method can finally get satisfied result.

关键词： Exaggerated portrait caricatures sequence-to-sequence model structure features

来源：评论

学校读者我要写书评

暂无评论

TACOTRON-BASED ACOUSTIC model USING PHONEME ALIGNMENT FOR PRACTICAL NEURAL TEXT-TO-SPEECH SYSTEMS

TACOTRON-BASED ACOUSTIC MODEL USING PHONEME ALIGNMENT FOR PR...

引用

IEEE Automatic Speech Recognition and Understanding Workshop (ASRU)

作者： Okamoto, Takuma Toda, Tomoki Shiga, Yoshinori Kawai, Hisashi Natl Inst Informat & Commun Technol Tokyo Japan Nagoya Univ Informat Technol Ctr Nagoya Aichi Japan

ISBN: (纸本)9781728103068

Although sequence-to-sequence (seq2seq) models with attention mechanism in neural text-to-speech (TTS) systems, such as Tacotron 2, can jointly optimize duration and acoustic models, and realize high-fidelity synthesis compared with conventional durationacoustic pipeline models, these involve a risk that speech samples cannot be sometimes successfully synthesized due to the attention prediction errors. Therefore, these seq2seq models cannot be directly introduced in practical TTS systems. On the other hand, the conventional pipeline models are broadly used in practical TTS systems since there are few crucial prediction errors in the duration model. For realizing high-quality practical TTS systems without attention prediction errors, this paper investigates Tacotron-based acoustic models with phoneme alignment instead of attention. The phoneme durations are first obtained from HMM-based forced alignment and the duration model is a simple bidirectional LSTM-based network. Then, a seq2seq model with forced alignment instead of attention is investigated and an alternative model with Tacotron decoder and phoneme duration is proposed. The results of experiments with full-context label input using WaveGlow vocoder indicate that the proposed model can realize a high-fidelity TTS system for Japanese with a real-time factor of 0.13 using a GPU without attention prediction errors compared with the seq2seq models.

关键词： Speech synthesis neural text-to-speech duration model forced alignment sequence-to-sequence model

来源：评论

学校读者我要写书评

暂无评论

sequence-to-sequence Imputation of Missing Sensor Data 32nd

Sequence-to-Sequence Imputation of Missing Sensor Data

引用

32nd Australasian Joint Conference on Artificial Intelligence

作者： Dabrowski, Joel Janek Rahman, Ashfaqur CSIRO Data61 Brisbane Qld Australia CSIRO Data61 Hobart Tas Australia

ISBN: (纸本)9783030352882;9783030352875

Although the sequence-to-sequence (encoder-decoder) model is considered the state-of-the-art in deep learning sequence models, there is little research into using this model for recovering missing sensor data. The key challenge is that the missing sensor data problem typically comprises three sequences (a sequence of observed samples, followed by a sequence of missing samples, followed by another sequence of observed samples) whereas, the sequence-to-sequence model only considers two sequences (an input sequence and an output sequence). We address this problem by formulating a sequence-to-sequence in a novel way. A forward RNN encodes the data observed before the missing sequence and a backward RNN encodes the data observed after the missing sequence. A decoder decodes the two encoders in a novel way to predict the missing data. We demonstrate that this model produces the lowest errors in 12% more cases than the current state-of-the-art.

关键词： Imputation Interpolation LSTM Encoder-decoder model sequence-to-sequence model

来源：评论

学校读者我要写书评

暂无评论

Personalized Dialogue Response Generation Learned from Monologues 20

Personalized Dialogue Response Generation Learned from Monol...

引用

Interspeech Conference

作者： Su, Feng-Guang Hsu, Aliyah R. Tuan, Yi-Lin Lee, Hung-Yi Natl Taiwan Univ Dept Elect Engn Taipei Taiwan

Personalized responses are essential for having an informative and human-like conversation. Because it is difficult to collect a large amount of dialogues involved with specific speakers, it is desirable that chatbot can learn to generate personalized responses simply from monologues of individuals. In this paper, we propose a novel personalized dialogue generation method which reduces the training data requirement to dialogues without speaker information and monologues of every target speaker. In the proposed approach, a generative adversarial network ensures the responses containing recognizable personal characteristics of the target speaker, and a backward SEQ2SEQ model reconstructs the input message for keeping the coherence of the generated responses. The proposed model demonstrates its flexibility to respond to open-domain conversations, and the experimental results show that the proposed method performs favorably against prior work in coherence, personality classification, and human evaluation.

关键词： personalized dialogue generation generative adversarial network sequence-to-sequence model monologues

来源：评论

学校读者我要写书评

暂无评论

Deep learning-based household electric energy consumption forecasting

引用

JOURNAL OF ENGINEERING-JOE 2020年第13期2020卷 639-642页

作者： Hyeon, Jonghwan Lee, HyeYoung Ko, Bowon Choi, Ho-Jin Korea Adv Inst Sci & Technol Sch Comp Daejeon South Korea

With the advent of various electronic products, the household electric energy consumption is continuously increasing, and therefore it becomes very important to predict the household electric energy consumption accurately. Energy prediction models also have been developed for decades with advanced machine learning technologies. Meanwhile, the deep learning models are still actively under study, and many newer models show the state-of-the-art performance. Therefore, it would be meaningful to conduct the same experiment with these new models. Here, the authors predict the household electric energy consumption using deep learning models, known to be suitable for dealing with time-series data. Specifically, vanilla long short-term memory (LSTM), sequence to sequence, and sequence to sequence with attention mechanism are used to predict the electric energy consumption in the household. As a result, the vanilla LSTM shows the best performance on the root-mean-square error metric. However, from a graphical point of view, it seems that the sequence-to-sequence model predicts the energy consumption patterns best and the vanilla LSTM does not follow the pattern well. Also, to achieve the best performance of each deep learning model, vanilla LSTM, sequence to sequence, and sequence to sequence with attention mechanism should observe past 72, 72, and 24 h, respectively.

关键词： power consumption time series learning (artificial intelligence) mean square error methods load forecasting power engineering computing recurrent neural nets vanilla LSTM deep learning model energy prediction models advanced machine learning technologies sequence-to-sequence model energy consumption patterns household electric energy consumption forecasting time-series data vanilla long short-term memory root-mean-square error metric sequence to sequence with attention mechanism time 72 0 hour time 24 0 hour

来源：评论

学校读者我要写书评

暂无评论

Robust sequence-to-sequence Acoustic modeling with Stepwise Monotonic Attention for Neural TTS 20

Robust Sequence-to-Sequence Acoustic Modeling with Stepwise ...

引用

Interspeech Conference

作者： He, Mutian Deng, Yan He, Lei Beihang Univ Sch Comp Sci & Engn Beijing Peoples R China Microsoft Beijing Peoples R China

Neural TTS has demonstrated strong capabilities to generate human-like speech with high quality and naturalness, while its generalization to out-of-domain texts is still a challenging task, with regard to the design of attention-based sequence-to-sequence acoustic modeling. Various errors occur in those inputs with unseen context, including attention collapse, skipping, repeating, etc., which limits the broader applications. In this paper, we propose a novel stepwise monotonic attention method in sequence-to-sequence acoustic modeling to improve the robustness on out-of-domain inputs. The method utilizes the strict monotonic property in TTS with constraints on monotonic hard attention that the alignments between inputs and outputs sequence must be not only monotonic but allowing no skipping on inputs. Soft attention could be used to evade mismatch between training and inference. The experimental results show that the proposed method could achieve significant improvements in robustness on out-of-domain scenarios for phoneme-based models, without any regression on the in-domain naturalness test.

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

来源：评论

学校读者我要写书评

暂无评论

ENHANCING HYBRID SELF-ATTENTION STRUCTURE WITH RELATIVE-POSITION-AWARE BIAS FOR SPEECH SYNTHESIS 44

ENHANCING HYBRID SELF-ATTENTION STRUCTURE WITH RELATIVE-POSI...

引用

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

作者： Yang, Shan Lu, Heng Kang, Shiying Xie, Lei Yu, Dong Northwestern Polytech Univ Sch Comp Sci Xian Shaanxi Peoples R China

ISBN: (纸本)9781479981311

Compared with the conventional "front-end"-"back-end"" vocoder" structure, based on the attention mechanism, end-to-end speech synthesis systems directly train and synthesize from text sequence to the acoustic feature sequence as a whole. Recently, a more calculation efficient end-to-end architecture named transformer, which is solely based on self-attention, was proposed to model global dependencies between the input and output sequences. However, although with many advantages, transformer lacks position information in its structure. Moreover, the weighted sum form in self-attention may disperse the attention to the whole input sequence other than focusing on the more important neighbouring positions. In order to solve the above problems, this paper introduces a hybrid self-attention structure which combines self-attention with the recurrent neural networks (RNNs). We further enhance the proposed structure with relative-position-aware biases. Mean opinion score (MOS) test results indicate that by enhancing hybrid self-attention structure with relative-position-aware biases, the proposed system achieves the best performance with only 0.11 MOS score lower than natural recording.

关键词： text-to-speech synthesis sequence-to-sequence model self-attention relative-position-aware representation

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：