检索结果-内蒙古大学图书馆

On using specaugment for end-to-end speech translation

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Bahar, Parnia Zeyer, Albert Schlüter, Ralf Ney, Hermann Human Language Technology and Pattern Recognition Group Computer Science Department RWTH Aachen University Aachen52062 Germany AppTek Aachen52062 Germany

This work investigates a simple data augmentation technique, SpecAugment, for end-to-end speech translation. SpecAugment is a low-cost implementation method applied directly to the audio input features and it consists of masking blocks of frequency channels, and/or time steps. We apply SpecAugment on end-to-end speech translation tasks and achieve up to +2.2% BLEU on LibriSpeech Audiobooks En→Fr and +1.2% on IWSLT TED-talks En→De by alleviating overfitting to some extent. We also examine the effectiveness of the method in a variety of data scenarios and show that the method also leads to significant improvements in various data conditions irrespective of the amount of training data. Copyright © 2019, The Authors. All rights reserved.

关键词：

LSTM language models for LVCSR in first-pass decoding and lattice-rescoring

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Beck, Eugen Zhou, Wei Schlüter, Ralf Ney, Hermann Human Language Technology and Pattern Recognition Computer Science Department RWTH Aachen University Aachen52074 Germany AppTek GmbH Aachen52062 Germany

LSTM based language models are an important part of modern LVCSR systems as they significantly improve performance over traditional backoff language models. Incorporating them efficiently into decoding has been notoriously difficult. In this paper we present an approach based on a combination of one-pass decoding and lattice rescoring. We perform decoding with the LSTM-LM in the first pass but recombine hypothesis that share the last two words, afterwards we rescore the resulting lattice. We run our systems on GPGPU equipped machines and are able to produce competitive results on the Hub5'00 and Librispeech evaluation corpora with a runtime better than real-time. In addition we shortly investigate the possibility to carry out the full sum over all state-sequences belonging to a given word-hypothesis during decoding without recombination. Copyright © 2019, The Authors. All rights reserved.

关键词： Decoding

Learning latent parameters without human response patterns: Item response theory with artificial crowds

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Lalor, John P. Wu, Hao Yu, Hong College of Information and Computer Sciences UMass Amherst Department of Psychology and Human Development Vanderbilt University Department of Computer Science UMass Lowell Center for Healthcare Organization and Implementation Research BedfordVA

Incorporating Item Response Theory (IRT) into NLP tasks can provide valuable information about model performance and behavior. Traditionally, IRT models are learned using human response pattern (RP) data, presenting a significant bottleneck for large data sets like those required for training deep neural networks (DNNs). In this work we propose learn-ing IRT models using RPs generated from arti-ficial crowds of DNN models. We demonstrate the effectiveness of learning IRT models using DNN-generated data through quantitative and qualitative analyses for two NLP tasks. Parameters learned from human and machine RPs for natural language inference and sentiment analysis exhibit medium to large positive cor-relations. We demonstrate a use-case for latent difficulty item parameters, namely training set filtering, and show that using difficulty to sample training data outperforms baseline methods. Finally, we highlight cases where human expectation about item difficulty does not match difficulty as estimated from the machine RPs. Copyright © 2019, The Authors. All rights reserved.

关键词： Deep neural networks

Detection of Astronauts’ Speech and language Disorder Signs during Space Missions using Natural language Processing Techniques

学校读者我要写书评

暂无评论

Detection of Astronauts’ Speech and Language Disorder Signs...

E-Health and Bioengineering Conference (EHB)

作者： Raluca Stefania Trofin Costin Chiru Cristian Vizitiu Adrian Dinculescu Ruxandra Vizitiu Alexandru Nistorescu Computer Science Department Politehnica University of Bucharest Romania Space Applications for Human Health and Safety Department Institute of Space Science Bucharest-Magurele Romania Ruxandra Vizitiu Speech and Language Clinic Bucharest Romania

As technology evolves at an alert pace, the desire and need for exploration of the outer space becomes more and more prevalent. human exploration and colonization of the Moon or Mars represents a strategic goal of international space agencies for the coming years. This implies a long-term exposure of astronauts in an extreme environment characterized by microgravity, isolation, confinement and other stressors which cause damage to both their bodies and minds. The longer the exposure to space environment, the greater the chance for an astronaut to face psychological problems. In, this regard, the herein paper is interdisciplinary orienting towards speech and language disorder analysis, especially on oral language disorders, for shaping a tool of behavioral psychology. Considering the well-known connection between speech and language disorders with interpersonal and intrapersonal psychological problems, this paper presents specific semantic assessments based on natural language processing techniques in order to gain understanding and shape a speech and language tool able to detect psychological impairments, especially for astronauts, but not merely.

关键词：

Hearing loss prevalence and years lived with disability, 1990-2019: Findings from the Global Burden of Disease Study 2019

学校读者我要写书评

暂无评论

The Lancet 2021年第10278期397卷 996-1009页

作者： Haile, Lydia M. Kamenov, Kaloyan Briant, Paul Svitil Orji, Aislyn U. Steinmetz, Jaimie D. Abdoli, Amir Abdollahi, Mohammad Abu-Gharbieh, Eman Afshin, Ashkan Ahmed, Haroon Rashid, Tarik Ahmed Akalu, Yonas Alahdab, Fares Alanezi, Fahad Mashhour Alanzi, Turki M. Al Hamad, Hanadi Ali, Liaqat Alipour, Vahid Al-Raddadi, Rajaa M. Amu, Hubert Arabloo, Jalal Arab-Zozani, Morteza Arulappan, Judie Ashbaugh, Charlie Atnafu, Desta Debalkie Babar, Zaheer-Ud-Din Baig, Atif Amin Banik, Palash Chandra Bärnighausen, Till Winfried Barrow, Amadou Bender, Rose G. Bhagavathula, Akshaya Srikanth Bhardwaj, Nikha Bhardwaj, Pankaj Bibi, Sadia Bijani, Ali Burkart, Katrin Cederroth, Christopher R. Charan, Jaykaran Choudhari, Sonali Gajanan Chu, Dinh-Toi Couto, Rosa A.S. Dagnew, Amare Belachew Dagnew, Baye Dahlawi, Saad M.A. Dai, Xiaochen Dandona, Lalit Dandona, Rakhi Desalew, Assefa Dhamnetiya, Deepak Dhimal, Mandira Lamichhane Dhimal, Meghnath Doyle, Kerrie E. Duncan, Bruce B. Ekholuenetale, Michael Filip, Irina Fischer, Florian Franklin, Richard Charles Gaidhane, Abhay Motiramji Gaidhane, Shilpa Gallus, Silvano Ghamari, Farhad Ghashghaee, Ahmad Ghozali, Ghozali Gilani, Syed Amir Glǎvan, Ionela-Roxana Golechha, Mahaveer Goulart, Barbara Niegia Garcia Gupta, Veer Bala Gupta, Vivek Kumar Hamidi, Samer Hammond, Billy Randall Hay, Simon I. Hayat, Khezar Heidari, Golnaz Hoffman, Howard J. Hopf, Kathleen Pillsbury Hosseinzadeh, Mehdi Househ, Mowafa Hussain, Rabia Hwang, Bing-Fang Iavicoli, Ivo Ibitoye, Segun Emmanuel Ilesanmi, Olayinka Stephen Irvani, Seyed Sina Naghibi Islam, Sheikh Mohammed Shariful Iwagami, Masao Jacob, Louis Jayapal, Sathish Kumar Jha, Ravi Prakash Jonas, Jost B. Kalhor, Rohollah Al-Salihi, Nawzad Kameran Kandel, Himal Kasa, Ayele Semachew Kayode, Gbenga A. Khalilov, Rovshan Khan, Ejaz Ahmad Khatib, Mahalaqua Nazli Kosen, Soewarta Koyanagi, Ai Kumar, G. Anil Landires, Ivan Lasrado, Savita Lim, Stephen S. Liu, Xuefeng Lobo, Stany W. Lugo, Alessandra Makki, Alaa Mendoza, Walter Mersha, Amanual Getnet Mihretie, Kebadnew Mulatu Institute for Health Metrics and Evaluation University of Washington Seattle WA United States Department of Noncommunicable Diseases World Health Organization (WHO) Geneva Switzerland Department of Parasitology and Mycology Jahrom University of Medical Sciences Jahrom Iran The Institute of Pharmaceutical Sciences (TIPS) Tehran University of Medical Sciences Tehran Iran School of Pharmacy Tehran University of Medical Sciences Tehran Iran Clinical Sciences Department University of Sharjah Sharjah United Arab Emirates Department of Health Metrics Sciences School of Medicine University of Washington Seattle WA United States Department of Biosciences COMSATS Institute of Information Technology Islamabad Pakistan Department of Computer Science and Engineering University of Kurdistan Hewler Erbil Iraq Department of Medical Physiology University of Gondar Gondar Ethiopia Mayo Evidence-based Practice Center Mayo Clinic Foundation for Medical Education and Research Rochester MN United States Imam Abdulrahman Bin Faisal University Dammam Saudi Arabia Health Information Management and Technology Department Dammam Saudi Arabia Geriatric and Long Term Care Department Hamad Medical Corporation Doha Qatar Rumailah Hospital Hamad Medical Corporation Doha Qatar Department of Biological Sciences National University of Medical Sciences (NUMS) Rawalpindi Pakistan Health Management and Economics Research Center Tehran Iran Department of Community Medicine King Abdulaziz University Jeddah Saudi Arabia Department of Population and Behavioural Sciences University of Health and Allied Sciences Ho Ghana Social Determinants of Health Research Center Birjand University of Medical Sciences Birjand Iran Department of Maternal and Child Health Sultan Qaboos University Muscat Oman Department of Health System and Health Economics Bahir Dar University Bahir Dar Ethiopia Department of Pharmacy University of Huddersfield Huddersfield United Kingdom Unit of Biochemistry Universiti

Background: Hearing loss affects access to spoken language, which can affect cognition and development, and can negatively affect social wellbeing. We present updated estimates from the Global Burden of Disease (GBD) study on the prevalence of hearing loss in 2019, as well as the condition's associated disability. Methods: We did systematic reviews of population-representative surveys on hearing loss prevalence from 1990 to 2019. We fitted nested meta-regression models for severity-specific prevalence, accounting for hearing aid coverage, cause, and the presence of tinnitus. We also forecasted the prevalence of hearing loss until 2050. Findings: An estimated 1·57 billion (95% uncertainty interval 1·51–1·64) people globally had hearing loss in 2019, accounting for one in five people (20·3% [19·5–21·1]). Of these, 403·3 million (357·3–449·5) people had hearing loss that was moderate or higher in severity after adjusting for hearing aid use, and 430·4 million (381·7–479·6) without adjustment. The largest number of people with moderate-to-complete hearing loss resided in the Western Pacific region (127·1 million people [112·3–142·6]). Of all people with a hearing impairment, 62·1% (60·2–63·9) were older than 50 years. The Healthcare Access and Quality (HAQ) Index explained 65·8% of the variation in national age-standardised rates of years lived with disability, because countries with a low HAQ Index had higher rates of years lived with disability. By 2050, a projected 2·45 billion (2·35–2·56) people will have hearing loss, a 56·1% (47·3–65·2) increase from 2019, despite stable age-standardised prevalence. Interpretation: As populations age, the number of people with hearing loss will increase. Interventions such as childhood screening, hearing aids, effective management of otitis media and meningitis, and cochlear implants have the potential to ameliorate this burden. Because the burden of moderate-to-complete hearing loss is concentrated in countries with low health-ca

关键词：

On using 2D sequence-to-sequence models for speech recognition

学校读者我要写书评

暂无评论

arXiv 2019年

Attention-based sequence-to-sequence models have shown promising results in automatic speech recognition. Using these architectures, one-dimensional input and output sequences are related by an attention approach, thereby replacing more explicit alignment processes, like in classical HMM-based modeling. In contrast, here we apply a novel two-dimensional long short-term memory (2DLSTM) architecture to directly model the input/output relation between audio/feature vector sequences and word sequences. The proposed model is an alternative model such that instead of using any type of attention components, we apply a 2DLSTM layer to assimilate the context from both input observations and output transcriptions. The experimental evaluation on the Switchboard 300h automatic speech recognition task shows word error rates for the 2DLSTM model that are competitive to end-to-end attention-based model. Copyright © 2019, The Authors. All rights reserved.

关键词： Speech recognition

language modeling with deep transformers

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Irie, Kazuki Zeyer, Albert Schlüter, Ralf Ney, Hermann Human Language Technology and Pattern Recognition Group Computer Science Department RWTH Aachen University Aachen52074 Germany AppTek GmbH Aachen52062 Germany

We explore deep autoregressive Transformer models in language modeling for speech recognition. We focus on two aspects. First, we revisit Transformer model configurations specifically for language modeling. We show that well configured Transformer models outperform our baseline models based on the shallow stack of LSTM recurrent neural network layers. We carry out experiments on the open-source LibriSpeech 960hr task, for both 200K vocabulary word-level and 10K byte-pair encoding subword-level language modeling. We apply our word-level models to conventional hybrid speech recognition by lattice rescoring, and the subword-level models to attention based encoder-decoder models by shallow fusion. Second, we show that deep Transformer language models do not require positional encoding. The positional encoding is an essential augmentation for the self-attention mechanism which is invariant to sequence ordering. However, in autoregressive setup, as is the case for language modeling, the amount of information increases along the position dimension, which is a positional signal by its own. The analysis of attention weights shows that deep autoregressive self-attention models can automatically make use of such positional information. We find that removing the positional encoding even slightly improves the performance of these models. Copyright © 2019, The Authors. All rights reserved.

关键词： Speech recognition

Genome-wide analyses identify 30 loci associated with obsessive–compulsive disorder

学校读者我要写书评

暂无评论

Nature Genetics 2025年 1-13页

作者： Strom, Nora I. Gerring, Zachary F. Galimberti, Marco Yu, Dongmei Halvorsen, Matthew W. Abdellaoui, Abdel Rodriguez-Fontenla, Cristina Sealock, Julia M. Bigdeli, Tim Coleman, Jonathan R. Mahjani, Behrang Thorp, Jackson G. Bey, Katharina Burton, Christie L. Luykx, Jurjen J. Zai, Gwyneth Alemany, Silvia Andre, Christine Askland, Kathleen D. Bäckman, Julia Banaj, Nerisa Barlassina, Cristina Nissen, Judith Becker Bienvenu, O. Joseph Black, Donald Bloch, Michael H. Børte, Sigrid Bosch, Rosa Breen, Michael Brennan, Brian P. Brentani, Helena Buxbaum, Joseph D. Bybjerg-Grauholm, Jonas Byrne, Enda M. Cabana-Dominguez, Judit Camarena, Beatriz Camarena, Adrian Cappi, Carolina Carracedo, Angel Casas, Miguel Cavallini, Maria Cristina Ciullo, Valentina Cook, Edwin H. Crosby, Jesse Cullen, Bernadette A. De Schipper, Elles J. Delorme, Richard Djurovic, Srdjan Elias, Jason A. Estivill, Xavier Falkenstein, Martha J. Fundin, Bengt T. Garner, Lauryn Gironda, Christina Goes, Fernando S. Grados, Marco A. Grove, Jakob Guo, Wei Haavik, Jan Hagen, Kristen Harrington, Kelly Havdahl, Alexandra Höffler, Kira D. Hounie, Ana G. Hucks, Donald Hultman, Christina Janecka, Magdalena Jenike, Eric Karlsson, Elinor K. Kelley, Kara Klawohn, Julia Krasnow, Janice E. Krebs, Kristi Lange, Christoph Lanzagorta, Nuria Levey, Daniel Lindblad-Toh, Kerstin Macciardi, Fabio Maher, Brion Mathes, Brittany McArthur, Evonne McGregor, Nathaniel McLaughlin, Nicole C. Meier, Sandra Miguel, Euripedes C. Mulhern, Maureen Nestadt, Paul S. Nurmi, Erika L. O’Connell, Kevin S. Osiecki, Lisa Ousdal, Olga Therese Palviainen, Teemu Pedersen, Nancy L. Piras, Federica Potluri, Sriramya Rabionet, Raquel Ramirez, Alfredo Rauch, Scott Reichenberg, Abraham Riddle, Mark A. Ripke, Stephan Rosário, Maria C. Sampaio, Aline S. Schiele, Miriam A. Skogholt, Anne Heidi Sloofman, Laura G. Smit, Jan Artigas, María Soler Thomas, Laurent F. Tifft, Eric Vallada, Homero van Kirk, Nathanial Veenstra-VanderWeele, Jeremy Vulink, Nienke N. Walker, Christopher P. Wang, Ying Wendland, Jens R. Winsvold Department of Psychology Humboldt-Universität zu Berlin Berlin Germany Department of Psychiatric Phenomics and Genomics (IPPG) Ludwig-Maximilians University Munich Munich Germany Centre for Psychiatry Research Department of Clinical Neuroscience Karolinska Institutet and Stockholm Health Services Region Stockholm Stockholm Sweden Department of Biomedicine Aarhus University Aarhus Denmark University Hospital of Psychiatry and Psychotherapy University of Bern Bern Switzerland Department of Mental Health and Neuroscience Translational Neurogenomics QIMR Berghofer Medical Research Institute Brisbane QLD Australia Department of Population Health and Immunity Healthy Development and Ageing Walter and Eliza Hall Institute of Medical Research Melbourne VIC Australia Department of Psychiatry Human Genetics Yale University New Haven CT United States VA Connecticut Healthcare System West Haven CT United States Department of Center for Genomic Medicine Psychiatric and Neurodevelopmental Genetics Unit Massachusetts General Hospital Boston MA United States Stanley Center for Psychiatric Research Broad Institute of MIT and Harvard Cambridge MA United States Department of Genetics University of North Carolina at Chapel Hill Chapel Hill NC United States Department of Psychiatry Amsterdam University Medical Centers University of Amsterdam Amsterdam Netherlands Center for Research in Molecular Medicine and Chronic Diseases (CiMUS) Genomics and Bioinformatics University of Santiago de Compostela Santiago de Compostela Spain Grupo de Medicina Xenómica Genetics Instituto de Investigación Sanitaria de Santiago de Compostela (FIDIS) Santiago de Compostela Spain Vanderbilt Genetics Institute Vanderbilt University Nashville TN United States Department of Psychiatry and Behavioral Sciences SUNY Downstate Health Sciences University Brooklyn NY United States VA NY Harbor Healthcare System Brooklyn NY United States Social Genetic and Developmental Psychiatry Centre Insti

Obsessive–compulsive disorder (OCD) affects ~1% of children and adults and is partly caused by genetic factors. We conducted a genome-wide association study (GWAS) meta-analysis combining 53,660 OCD cases and 2,044,417 controls and identified 30 independent genome-wide significant loci. Gene-based approaches identified 249 potential effector genes for OCD, with 25 of these classified as the most likely causal candidates, including WDR6, DALRD3 and CTNND1 and multiple genes in the major histocompatibility complex (MHC) region. We estimated that ~11,500 genetic variants explained 90% of OCD genetic heritability. OCD genetic risk was associated with excitatory neurons in the hippocampus and the cortex, along with D₁ and D₂ type dopamine receptor-containing medium spiny neurons. OCD genetic risk was shared with 65 of 112 additional phenotypes, including all the psychiatric disorders we examined. In particular, OCD shared genetic risk with anxiety, depression, anorexia nervosa and Tourette syndrome and was negatively associated with inflammatory bowel diseases, educational attainment and body mass index. © The Author(s) 2025.

关键词：

The impact of human dispersals and local interactions on the genetic diversity of coastal Papua New Guinea over the past 2,500 years

学校读者我要写书评

暂无评论

Nature Ecology & Evolution 2025年 1-16页

作者： Kathrin Nägele Adam B. Rohrlach Selina Carlhoff Yilei Huang Harald Ringbauer Emilie Bertolini Rita Radzeviciute Cosimo Posth Johannes Krause Rebecca Kinaston Dylan Gaffney Monica Tromp Dimitri Anson Glenn Summerhayes Mary Walworth Fiona Petchey Peter Petchey Claudine Stirling Malcolm Reid David Barr Ben Shaw Hallie Buckley Adam Powell Department of Archaeogenetics Max Planck Institute for Evolutionary Anthropology Leipzig Germany School of Computer and Mathematical Sciences University of Adelaide Adelaide South Australia Australia Bioinformatics Group Institute of Computer Science Universität Leipzig Leipzig Germany Department of Human Evolutionary Biology Harvard University Cambridge MA USA Archaeo- and Palaeogenetics Institute for Archaeological Sciences Department of Geosciences University of Tübingen Tübingen Germany Senckenberg Centre for Human Evolution and Palaeoenvironment University of Tübingen Tübingen Germany BioArch South Waitati New Zealand School of Archaeology University of Oxford Oxford UK Archaeology Programme University of Otago Dunedin New Zealand Southern Pacific Archaeological Research (SPAR) University of Otago Dunedin New Zealand School of Social Science University of Queensland Saint Lucia Queensland Australia Department of Linguistic and Cultural Evolution Max-Planck-Institute for Evolutionary Anthropology Leipzig Germany CRLAO (UMR8563) Centre National de la Recherche Scientifique Paris France Radiocarbon Dating Laboratory Te Aka Mātuatua – School of Science University of Waikato Hamilton New Zealand ARC Centre of Excellence for Australian Biodiversity and Heritage College of Arts Society and Education James Cook University Cairns Queensland Australia Southern Archaeology Ltd Dunedin New Zealand Department of Geology University of Otago Dunedin New Zealand Centre for Trace Element Analysis Department of Geology University of Otago Dunedin New Zealand School of Culture History and Language Australian National University Canberra Australian Capital Territory Australia Department of Anatomy School of Biomedical Sciences University of Otago Dunedin New Zealand Department of Human Behavior Ecology and Culture Max Planck Institute for Evolutionary Anthropology Leipzig Germany

The inhabitants of New Guinea and its outlying islands have played an important role in the human history of the Pacific region. Nevertheless, the genetic diversity, particularly of pre-colonial communities, is still understudied. Here we present the ancient genomes of 42 individuals from Papua New Guinea (PNG). The ancient genomic results of individuals from Watom Island (Bismarck Archipelago) and the south and northeastern coasts of PNG are contextualized with new (bio-) archaeological data. The individuals’ accelerator mass spectrometry (AMS) dates span 2,500 years of human habitation, and our results demonstrate the influences of different dispersal events on the genetic make-up of ancient PNG communities. The oldest individuals show an unadmixed Papuan-related genetic signature, whereas individuals dating from 2,100 years before present carry varying degrees of an East-Asian-related contribution. These results and the inferred admixture dates suggest a centuries-long delay in genetic mixture with local communities after the arrival of populations with Asian ancestry. Two geographically close communities on the South Coast, AMS dated to within the past 540 years, diverge in their genetic profiles, suggesting differences in their interaction spheres involving groups with distinct ancestries. The inferred split time of these communities around 650 years before present coincides with intensified settlement activity and the emergence of regional trade networks.

关键词：