Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems
For purposes of automated speech recognition in under-resourced environments, techniques used to share acoustic data between closely related or similar languages become important. Donor languages with abundant resources can potentially be used to increase the recognition accuracy of speech systems developed in the resource poor target language. The assumption is that adding more data will increase the robustness of the statistical estimations captured by the acoustic models. In this study we investigated data sharing between Afrikaans and Flemish - an under-resourced and well-resourced language, respectively. Our approach was focused on the exploration of model adaptation and refinement techniques associated with hidden Markov model based speech recognition systems to improve the benefit of sharing data. Specifically, we focused on the use of currently available techniques, some possible combinations and the exact utilisation of the techniques during the acoustic model development process. Our findings show that simply using normal approaches to adaptation and refinement does not result in any benefits when adding Flemish data to the Afrikaans training pool. The only observed improvement was achieved when developing acoustic models on all available data but estimating model refinements and adaptations on the target data only. SIGNIFICANCE: Acoustic modelling for under-resourced languages Automatic speech recognition for Afrikaans Data sharing between Flemish and Afrikaans to improve acoustic modelling for Afrikaans
Main Authors: | , , , |
---|---|
Format: | Digital revista |
Language: | English |
Published: |
Academy of Science of South Africa
2017
|
Online Access: | http://www.scielo.org.za/scielo.php?script=sci_arttext&pid=S0038-23532017000100009 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
id |
oai:scielo:S0038-23532017000100009 |
---|---|
record_format |
ojs |
spelling |
oai:scielo:S0038-235320170001000092017-02-23Speech recognition for under-resourced languages: Data sharing in hidden Markov model systemsde Wet,FebeKleynhans,Neilvan Compernolle,DirkSahraeian,Reza acoustic modelling Afrikaans Flemish automatic speech recognition For purposes of automated speech recognition in under-resourced environments, techniques used to share acoustic data between closely related or similar languages become important. Donor languages with abundant resources can potentially be used to increase the recognition accuracy of speech systems developed in the resource poor target language. The assumption is that adding more data will increase the robustness of the statistical estimations captured by the acoustic models. In this study we investigated data sharing between Afrikaans and Flemish - an under-resourced and well-resourced language, respectively. Our approach was focused on the exploration of model adaptation and refinement techniques associated with hidden Markov model based speech recognition systems to improve the benefit of sharing data. Specifically, we focused on the use of currently available techniques, some possible combinations and the exact utilisation of the techniques during the acoustic model development process. Our findings show that simply using normal approaches to adaptation and refinement does not result in any benefits when adding Flemish data to the Afrikaans training pool. The only observed improvement was achieved when developing acoustic models on all available data but estimating model refinements and adaptations on the target data only. SIGNIFICANCE: Acoustic modelling for under-resourced languages Automatic speech recognition for Afrikaans Data sharing between Flemish and Afrikaans to improve acoustic modelling for AfrikaansAcademy of Science of South AfricaSouth African Journal of Science v.113 n.1-2 20172017-02-01journal articletext/htmlhttp://www.scielo.org.za/scielo.php?script=sci_arttext&pid=S0038-23532017000100009en |
institution |
SCIELO |
collection |
OJS |
country |
Sudáfrica |
countrycode |
ZA |
component |
Revista |
access |
En linea |
databasecode |
rev-scielo-za |
tag |
revista |
region |
África del Sur |
libraryname |
SciELO |
language |
English |
format |
Digital |
author |
de Wet,Febe Kleynhans,Neil van Compernolle,Dirk Sahraeian,Reza |
spellingShingle |
de Wet,Febe Kleynhans,Neil van Compernolle,Dirk Sahraeian,Reza Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
author_facet |
de Wet,Febe Kleynhans,Neil van Compernolle,Dirk Sahraeian,Reza |
author_sort |
de Wet,Febe |
title |
Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
title_short |
Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
title_full |
Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
title_fullStr |
Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
title_full_unstemmed |
Speech recognition for under-resourced languages: Data sharing in hidden Markov model systems |
title_sort |
speech recognition for under-resourced languages: data sharing in hidden markov model systems |
description |
For purposes of automated speech recognition in under-resourced environments, techniques used to share acoustic data between closely related or similar languages become important. Donor languages with abundant resources can potentially be used to increase the recognition accuracy of speech systems developed in the resource poor target language. The assumption is that adding more data will increase the robustness of the statistical estimations captured by the acoustic models. In this study we investigated data sharing between Afrikaans and Flemish - an under-resourced and well-resourced language, respectively. Our approach was focused on the exploration of model adaptation and refinement techniques associated with hidden Markov model based speech recognition systems to improve the benefit of sharing data. Specifically, we focused on the use of currently available techniques, some possible combinations and the exact utilisation of the techniques during the acoustic model development process. Our findings show that simply using normal approaches to adaptation and refinement does not result in any benefits when adding Flemish data to the Afrikaans training pool. The only observed improvement was achieved when developing acoustic models on all available data but estimating model refinements and adaptations on the target data only. SIGNIFICANCE: Acoustic modelling for under-resourced languages Automatic speech recognition for Afrikaans Data sharing between Flemish and Afrikaans to improve acoustic modelling for Afrikaans |
publisher |
Academy of Science of South Africa |
publishDate |
2017 |
url |
http://www.scielo.org.za/scielo.php?script=sci_arttext&pid=S0038-23532017000100009 |
work_keys_str_mv |
AT dewetfebe speechrecognitionforunderresourcedlanguagesdatasharinginhiddenmarkovmodelsystems AT kleynhansneil speechrecognitionforunderresourcedlanguagesdatasharinginhiddenmarkovmodelsystems AT vancompernolledirk speechrecognitionforunderresourcedlanguagesdatasharinginhiddenmarkovmodelsystems AT sahraeianreza speechrecognitionforunderresourcedlanguagesdatasharinginhiddenmarkovmodelsystems |
_version_ |
1756004837157240832 |