1/2015 - 9 |
Enhancing ASR Systems for Under-Resourced Languages through a Novel Unsupervised Acoustic Model Training TechniqueCUCU, H. , BUZO, A. , BESACIER, L. , BURILEANU, C. |
View the paper record and citations in |
Click to see author's profile in SCOPUS, IEEE Xplore, Web of Science |
Download PDF (612 KB) | Citation | Downloads: 913 | Views: 4,169 |
Author keywords
speech recognition, under-resourced languages, unsupervised acoustic modeling, unsupervised training
References keywords
speech(15), training(13), unsupervised(12), resourced(5), recognition(5), processing(5), languages(5), language(5), acoustic(5), system(4)
Blue keywords are present in both the references section and the paper title.
About this article
Date of Publication: 2015-02-28
Volume 15, Issue 1, Year 2015, On page(s): 63 - 68
ISSN: 1582-7445, e-ISSN: 1844-7600
Digital Object Identifier: 10.4316/AECE.2015.01009
Web of Science Accession Number: 000352158600009
SCOPUS ID: 84924787729
Abstract
Statistical speech and language processing techniques, requiring large amounts of training data, are currently state-of-the-art in automatic speech recognition. For high-resourced, international languages this data is widely available, while for under-resourced languages the lack of data poses serious problems. Unsupervised acoustic modeling can offer a cost and time effective way of creating a solid acoustic model for any under-resourced language. This study describes a novel unsupervised acoustic model training method and evaluates it on speech data in an under-resourced language: Romanian. The key novel factor of the method is the usage of two complementary seed ASR systems to produce high quality transcriptions, with a Character Error Rate (ChER) < 5%, for initially untranscribed speech data. The methodology leads to a relative Word Error Rate (WER) improvement of more than 10% when 100 hours of untranscribed speech are used. |
References | | | Cited By |
Web of Science® Times Cited: 4 [View]
View record in Web of Science® [View]
View Related Records® [View]
Updated 2 days, 14 hours ago
SCOPUS® Times Cited: 5
View record in SCOPUS® [Free preview]
View citations in SCOPUS® [Free preview]
[1] Semi-Supervised Training of Language Model on Spanish Conversational Telephone Speech Data, Egorova, Ekaterina, Serrano, Jordi Luque, Procedia Computer Science, ISSN 1877-0509, Issue , 2016.
Digital Object Identifier: 10.1016/j.procs.2016.04.038 [CrossRef]
[2] Progress on automatic annotation of speech corpora using complementary ASR systems, Georgescu, Alexandru-Lucian, Cucu, Horia, Burileanu, Corneliu, 2019 42nd International Conference on Telecommunications and Signal Processing (TSP), ISBN 978-1-7281-1864-2, 2019.
Digital Object Identifier: 10.1109/TSP.2019.8769087 [CrossRef]
[3] Automatic Annotation of Speech Corpora using Approximate Transcripts, Manolache, Cristian, Georgescu, Alexandru-Lucian, Caranica, Alexandru, Cucu, Horia, 2020 43rd International Conference on Telecommunications and Signal Processing (TSP), ISBN 978-1-7281-6376-5, 2020.
Digital Object Identifier: 10.1109/TSP49548.2020.9163405 [CrossRef]
[4] Automatic Annotation of Speech Corpora Using Complementary GMM and DNN Acoustic Models, Georgescu, Alexandru-Lucian, Cucu, Horia, 2018 41st International Conference on Telecommunications and Signal Processing (TSP), ISBN 978-1-5386-4695-3, 2018.
Digital Object Identifier: 10.1109/TSP.2018.8441374 [CrossRef]
[5] Data-Filtering Methods for Self-Training of Automatic Speech Recognition Systems, Georgescu, Alexandru-Lucian, Manolache, Cristian, Oneata, Dan, Cucu, Horia, Burileanu, Corneliu, 2021 IEEE Spoken Language Technology Workshop (SLT), ISBN 978-1-7281-7066-4, 2021.
Digital Object Identifier: 10.1109/SLT48900.2021.9383577 [CrossRef]
Disclaimer: All information displayed above was retrieved by using remote connections to respective databases. For the best user experience, we update all data by using background processes, and use caches in order to reduce the load on the servers we retrieve the information from. As we have no control on the availability of the database servers and sometimes the Internet connectivity may be affected, we do not guarantee the information is correct or complete. For the most accurate data, please always consult the database sites directly. Some external links require authentication or an institutional subscription.
Web of Science® is a registered trademark of Clarivate Analytics, Scopus® is a registered trademark of Elsevier B.V., other product names, company names, brand names, trademarks and logos are the property of their respective owners.
Faculty of Electrical Engineering and Computer Science
Stefan cel Mare University of Suceava, Romania
All rights reserved: Advances in Electrical and Computer Engineering is a registered trademark of the Stefan cel Mare University of Suceava. No part of this publication may be reproduced, stored in a retrieval system, photocopied, recorded or archived, without the written permission from the Editor. When authors submit their papers for publication, they agree that the copyright for their article be transferred to the Faculty of Electrical Engineering and Computer Science, Stefan cel Mare University of Suceava, Romania, if and only if the articles are accepted for publication. The copyright covers the exclusive rights to reproduce and distribute the article, including reprints and translations.
Permission for other use: The copyright owner's consent does not extend to copying for general distribution, for promotion, for creating new works, or for resale. Specific written permission must be obtained from the Editor for such copying. Direct linking to files hosted on this website is strictly prohibited.
Disclaimer: Whilst every effort is made by the publishers and editorial board to see that no inaccurate or misleading data, opinions or statements appear in this journal, they wish to make it clear that all information and opinions formulated in the articles, as well as linguistic accuracy, are the sole responsibility of the author.