检索结果-内蒙古大学图书馆

A GLANCE IS ENOUGH: EXTRACT TARGET SENTENCE BY LOOKING AT A KEYWORD

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Shi, Ying Wang, Dong Li, Lantian Han, Jiqing School of Computer Science and Technology Harbin Institute of Technology China Center for Speech and Language Technologies BNRist Tsinghua University China School of Artificial Intelligence Beijing University of Posts and Telecommunications China

This paper investigates the possibility of extracting a target sentence from multi-talker speech using only a keyword as input. For example, in social security applications, the keyword might be "help", and the goal is to identify what the person who called for help is articulating while ignoring other speakers. To address this problem, we propose using the Transformer architecture to embed both the keyword and the speech utterance and then rely on the cross-attention mechanism to select the correct content from the concatenated or overlapping speech. Experimental results on Librispeech demonstrate that our proposed method can effectively extract target sentences from very noisy and mixed speech (SNR=-3dB), achieving a phone error rate (PER) of 26%, compared to the baseline system's PER of 96%. Copyright © 2023, The Authors. All rights reserved.

关键词： speech recognition

Distributed Dual Averaging for Composite Optimization

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Wu, Ping Lu, Haolin Liu, Zhengyang Heyan, Huang School of Computer Science and Technology Beijing Institute of Technology Beijing100081 China Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications China

In the context of a multi-agent network, we investigate the distributed online composite optimization problem where the optimization objective of each local agent is a combination of a convex cost function and a convex regularized function. To handle this problem, we introduce two distributed algorithms based on approximate dual averaging. The first online distributed composite dual averaging algorithm, under the gradient feedback, achieves a composite regret bound of $\mathcal{O}(\sqrt{T})$ and $\mathcal{O}(\ln T)$ for general convex and strongly convex regularized functions, respectively, across a total time horizon of $T$. Furthermore, the second one, relying exclusively on the values of cost function rather than the full information of gradient, attains a comparable composite regret bound against the first algorithm. To illustrate the effectiveness of our proposed algorithms, we present a plethora of numerical simulations for both the distributed regularized regression task and the distributed regularized classification task. © 2023, The Authors. All rights reserved.

关键词： Cost functions

Dynamic Regret with Unknown Delays

学校读者我要写书评

暂无评论

SSRN

SSRN 2023年

作者： Wu, Ping Liu, Zhengyang Lu, Haolin Huang, Heyan School of Computer Science and Technology Beijing Institute of Technology Beijing100081 China Beijing Engineering Research Center of High Volume Language Information Processing and Cloud Computing Applications China

In this paper, we delve into the realm of delayed online convex optimization, addressing scenarios where unknown delays exist between the querying and revealing of feedback. Obtaining dynamic regret in this case poses challenges arising from the uncertainties associated with both comparator sequence and delay sequence. First, to handle the uncertainty of comparator sequence, we introduce an adaptive delayed algorithm, yielding an $\mathcal{O}(\max{\sqrt{(1+\Delta_T)D_T},(\hat{d}-1)\Delta_T})$ dynamic regret bound for arbitrary comparator sequences. Here, $\Delta_T$ represents the path-length of the comparator sequence, $\hat{d}$ denotes the maximum delay, and $D_T$ signifies the cumulative delay over the entire horizon. Our approach centers around employing a delayed meta algorithm to track parallel delayed expert algorithms with varying step sizes. To complement our theoretical analysis, we establish an $\Omega(\sqrt{(1+\Delta_T)D_T})$ lower bound. Furthermore, to address the uncertainty of delay sequence, we extend our adaptive algorithm to scenarios where the time horizon and cumulative delay are unknown, incorporating a delayed doubling trick tailored to the number of missing feedback instances, thereby achieving a comparable dynamic regret. Finally, numerical simulations are presented to validate the effectiveness of our proposed approach. © 2023, The Authors. All rights reserved.

关键词： Comparators (optical)

MADNet: Maximizing Addressee Deduction Expectation for Multi-Party Conversation Generation

学校读者我要写书评

暂无评论

arXiv 2023年

作者： Gu, Jia-Chen Tan, Chao-Hong Chu, Caiyuan Ling, Zhen-Hua Tao, Chongyang Liu, Quan Liu, Cong National Engineering Research Center of Speech and Language Information Processing University of Science and Technology of China Hefei China iFLYTEK Research Hefei China Peking University Beijing China State Key Laboratory of Cognitive Intelligence China

Modeling multi-party conversations (MPCs) with graph neural networks has been proven effective at capturing complicated and graphical information flows. However, existing methods rely heavily on the necessary addressee labels and can only be applied to an ideal setting where each utterance must be tagged with an "@" or other equivalent addressee label. To study the scarcity of addressee labels which is a common issue in MPCs, we propose MADNet that maximizes addressee deduction expectation in heterogeneous graph neural networks for MPC generation. Given an MPC with a few addressee labels missing, existing methods fail to build a consecutively connected conversation graph, but only a few separate conversation fragments instead. To ensure message passing between these conversation fragments, four additional types of latent edges are designed to complete a fully-connected graph. Besides, to optimize the edge-type-dependent message passing for those utterances without addressee labels, an Expectation-Maximization-based method that iteratively generates silver addressee labels (E step), and optimizes the quality of generated responses (M step), is designed. Experimental results on two Ubuntu IRC channel benchmarks show that MADNet outperforms various baseline models on the task of MPC generation, especially under the more common and challenging setting where part of addressee labels are missing. Copyright © 2023, The Authors. All rights reserved.

关键词： Message passing

Blockchain-Based Framework for Stock Market Operations: IPOs, Trading, and Dividend Distribution

学校读者我要写书评

暂无评论

Blockchain-Based Framework for Stock Market Operations: IPOs...

IEEE International Conference on Cloud Computing Technology and science (CloudCom)

作者： Muhammad Sharshar Kareem Elzeky Menna Othman SeifAldin Khaled Hosam Moubarak Walid Gomaa Department of Computer Science Center for Secure Cyber-Physical Systems (C2PS) Khalifa University UAE Department of Natural Language Processing MBZUAI UAE Department of Computer Engineering Cyber-Physical Lab Egypt-Japan University of Science and Technology Egypt Accounting Department Egypt-Japan University of Science and Technology Egypt

ISBN: (数字)9798331507589

ISBN: (纸本)9798331507596

This study examines blockchain technology's potential to enhance the stock market by offering a decentralized alternative to the traditional system. The study explores the challenges of the current traditional stock market, such as long settlement times, weak transparency, dependency on centralized clearing houses, and heightened fees. The proposed novel system leverages blockchain technology to create a fully decentralized stock market encompassing all fundamental operations from Initial Public Offerings (IPOs) to stock trading and extending to distributing profits or dividends. It connects all market participants through an Ethereum consortium blockchain fuelled by smart contracts. This approach replaces traditional intermediaries like brokers with automated regulations, ensuring more rapid transaction settlements, reduced operational errors and fees, and strengthened security measures. The proposed blockchain-based system, consisting of four smart contracts for IPO, stock exchange, dividend distribution, and participant management, demonstrates remarkable performance, showing promise in efficiently handling the current load of the largest stock markets like the London Stock Exchange (LSE). Through theoretical and practical testing, the system achieved a remarkable throughput of 120 transactions per second and latency or finality of just 4 seconds, underscoring its capability to significantly enhance the efficiency of stock market operations. 1 1 The implementation code is available via GitHub upon request.

关键词： Cloud computing Smart contracts Transforms Throughput Regulation Blockchains Security Stock markets Testing Software development management

VITS-Based Singing Voice Conversion System with DSPGAN Post-processing for SVCC2023

学校读者我要写书评

暂无评论

VITS-Based Singing Voice Conversion System with DSPGAN Post-...

IEEE Workshop on Automatic speech Recognition and Understanding

作者： Yiquan Zhou Meng Chen Yi Lei Jihua Zhu Weifeng Zhao School of Software Engineering Xi’an Jiaotong University Xi’an China Audio Speech and Language Processing Group (ASLP@NPU) School of Computer Science Northwestern Polytechnical University Xi’an China Lyra Lab Tencent Music Entertainment Shenzhen China

This paper presents the T02 team’s system for the Singing Voice Conversion Challenge 2023 (SVCC2023). Our system entails a VITS-based SVC model, incorporating three modules: a feature extractor, a voice converter, and a postprocessor. Specifically, the feature extractor provides F0 contours and extracts speaker-independent linguistic content from the input singing voice by leveraging a HuBERT model. The voice converter is employed to recompose the speaker timbre, F0, and linguistic content to generate the waveform of the target speaker. Besides, to further improve the audio quality, a fine-tuned DSPGAN vocoder is introduced to resynthesise the waveform. Given the limited target speaker data, we utilize a two-stage training strategy to adapt the base model to the target speaker. During model adaptation, several tricks, such as data augmentation and joint training with auxiliary singer data, are involved. Official challenge results show that our system achieves superior performance, especially in the cross-domain task, ranking 1st and 2nd in naturalness and similarity, respectively. Further ablation justifies the effectiveness of our system design.

关键词：

Dualvc 2: Dynamic Masked Convolution for Unified Streaming and Non-Streaming Voice Conversion

学校读者我要写书评

暂无评论

Dualvc 2: Dynamic Masked Convolution for Unified Streaming a...

International Conference on Acoustics, speech, and Signal processing (ICASSP)

作者： Ziqian Ning Yuepeng Jiang Pengcheng Zhu Shuai Wang Jixun Yao Lei Xie Mengxiao Bi Audio Speech and Language Processing Group (ASLP@NPU) School of Computer Science Northwestern Polytechnical University Xi’an China Fuxi AI Lab NetEase Inc. Hangzhou China Shenzhen Research Institute of Big Data The Chinese University of Hong Kong Shenzhen (CUHK-Shenzhen) China

Voice conversion is becoming increasingly popular, and a growing number of application scenarios require models with streaming inference capabilities. The recently proposed DualVC attempts to achieve this objective through streaming model architecture design and intra-model knowledge distillation along with hybrid predictive coding to compensate for the lack of future information. However, DualVC encounters several problems that limit its performance. First, the autoregressive decoder has error accumulation in its nature and limits the inference speed as well. Second, the causal convolution enables streaming capability but cannot sufficiently use future information within chunks. Third, the model is unable to effectively address the noise in the unvoiced segments, lowering the sound quality. In this paper, we propose DualVC 2 to address these issues. Specifically, the model backbone is migrated to a Conformer-based architecture, empowering parallel inference. Causal convolution is replaced by non-causal convolution with a dynamic chunk mask to make better use of within-chunk future information. Also, quiet attention is introduced to enhance the model’s noise robustness. Experiments show that DualVC 2 outperforms DualVC and other baseline systems in both subjective and objective metrics, with only 186.4 ms latency. Our audio samples are made publicly available 1 .

关键词：

LONGEMBED: Extending Embedding Models for Long Context Retrieval

学校读者我要写书评

暂无评论

arXiv 2024年

作者： Zhu, Dawei Wang, Liang Yang, Nan Song, Yifan Wu, Wenhao Wei, Furu Li, Sujian School of Computer Science Peking University China National Key Laboratory for Multimedia Information Processing Peking University China Jiangsu Collaborative Innovation Center for Language Ability Jiangsu Normal University China Microsoft Corporation United States

Embedding models play a pivotal role in modern NLP applications such as document retrieval. However, existing embedding models are limited to encoding short documents of typically 512 tokens, restrained from application scenarios requiring long inputs. This paper explores context window extension of existing embedding models, pushing their input length to a maximum of 32,768. We begin by evaluating the performance of existing embedding models using our newly constructed LONGEMBED benchmark, which includes two synthetic and four real-world tasks, featuring documents of varying lengths and dispersed target information. The benchmarking results highlight huge opportunities for enhancement in current models. Via comprehensive experiments, we demonstrate that training-free context window extension strategies can effectively increase the input length of these models by several folds. Moreover, comparison of models using Absolute Position Encoding (APE) and Rotary Position Encoding (RoPE) reveals the superiority of RoPE-based embedding models in context window extension, offering empirical guidance for future models. Our benchmark, code and trained models will be released to advance the research in long context embedding models. © 2024, CC BY.

关键词： Rope

Generating Sequences by Learning to Self-Correct

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Welleck, Sean Lu, Ximing West, Peter Brahman, Faeze Shen, Tianxiao Khashabi, Daniel Choi, Yejin Allen Institute for Artificial Intelligence United States Center for Language and Speech Processing Johns Hopkins University United States Paul G. Allen School of Computer Science & Engineering University of Washington United States

Sequence generation applications require satisfying semantic constraints, such as ensuring that programs are correct, using certain keywords, or avoiding undesirable content. language models, whether fine-tuned or prompted with few-shot demonstrations, frequently violate these constraints, and lack a mechanism to iteratively revise their outputs. Moreover, some powerful language models are of extreme scale or inaccessible, making it inefficient, if not infeasible, to update their parameters for task-specific adaptation. We present SELF-CORRECTION, an approach that decouples an imperfect base generator (an off-the-shelf language model or supervised sequence-to-sequence model) from a separate corrector that learns to iteratively correct imperfect generations. To train the corrector, we propose an online training procedure that can use either scalar or natural language feedback on intermediate imperfect generations. We show that SELF-CORRECTION improves upon the base generator in three diverse generation tasks-mathematical program synthesis, lexically-constrained generation, and toxicity control- even when the corrector is much smaller than the base generator. © 2022, CC BY.

关键词： Iterative methods