Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case
Recently, there has been a growth in the number of studies addressing the automatic processing of low-resource languages. The lack of speech and text data significantly hinders the development of speech technologies for such languages. This paper introduces an automatic speech recognition system for...
Published in: | Mathematics |
---|---|
Main Authors: | , |
Format: | Text |
Language: | English |
Published: |
Multidisciplinary Digital Publishing Institute
2023
|
Subjects: | |
Online Access: | https://doi.org/10.3390/math11183814 |
id |
ftmdpi:oai:mdpi.com:/2227-7390/11/18/3814/ |
---|---|
record_format |
openpolar |
spelling |
ftmdpi:oai:mdpi.com:/2227-7390/11/18/3814/ 2023-10-09T21:53:08+02:00 Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case Irina Kipyatkova Ildar Kagirov 2023-09-05 application/pdf https://doi.org/10.3390/math11183814 eng eng Multidisciplinary Digital Publishing Institute Network Science https://dx.doi.org/10.3390/math11183814 https://creativecommons.org/licenses/by/4.0/ Mathematics Volume 11 Issue 18 Pages: 3814 low-resource languages automatic speech recognition audio data augmentation time delay neural network hidden Markov models long short-term memory Text 2023 ftmdpi https://doi.org/10.3390/math11183814 2023-09-10T23:54:15Z Recently, there has been a growth in the number of studies addressing the automatic processing of low-resource languages. The lack of speech and text data significantly hinders the development of speech technologies for such languages. This paper introduces an automatic speech recognition system for Livvi-Karelian. Acoustic models based on artificial neural networks with time delays and hidden Markov models were trained using a limited speech dataset of 3.5 h. To augment the data, pitch and speech rate perturbation, SpecAugment, and their combinations were employed. Language models based on 3-grams and neural networks were trained using written texts and transcripts. The achieved word error rate metric of 22.80% is comparable to other low-resource languages. To the best of our knowledge, this is the first speech recognition system for Livvi-Karelian. The results obtained can be of a certain significance for development of automatic speech recognition systems not only for Livvi-Karelian, but also for other low-resource languages, including the fields of speech recognition and machine translation systems. Future work includes experiments with Karelian data using techniques such as transfer learning and DNN language models. Text karelian MDPI Open Access Publishing Mathematics 11 18 3814 |
institution |
Open Polar |
collection |
MDPI Open Access Publishing |
op_collection_id |
ftmdpi |
language |
English |
topic |
low-resource languages automatic speech recognition audio data augmentation time delay neural network hidden Markov models long short-term memory |
spellingShingle |
low-resource languages automatic speech recognition audio data augmentation time delay neural network hidden Markov models long short-term memory Irina Kipyatkova Ildar Kagirov Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
topic_facet |
low-resource languages automatic speech recognition audio data augmentation time delay neural network hidden Markov models long short-term memory |
description |
Recently, there has been a growth in the number of studies addressing the automatic processing of low-resource languages. The lack of speech and text data significantly hinders the development of speech technologies for such languages. This paper introduces an automatic speech recognition system for Livvi-Karelian. Acoustic models based on artificial neural networks with time delays and hidden Markov models were trained using a limited speech dataset of 3.5 h. To augment the data, pitch and speech rate perturbation, SpecAugment, and their combinations were employed. Language models based on 3-grams and neural networks were trained using written texts and transcripts. The achieved word error rate metric of 22.80% is comparable to other low-resource languages. To the best of our knowledge, this is the first speech recognition system for Livvi-Karelian. The results obtained can be of a certain significance for development of automatic speech recognition systems not only for Livvi-Karelian, but also for other low-resource languages, including the fields of speech recognition and machine translation systems. Future work includes experiments with Karelian data using techniques such as transfer learning and DNN language models. |
format |
Text |
author |
Irina Kipyatkova Ildar Kagirov |
author_facet |
Irina Kipyatkova Ildar Kagirov |
author_sort |
Irina Kipyatkova |
title |
Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
title_short |
Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
title_full |
Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
title_fullStr |
Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
title_full_unstemmed |
Deep Models for Low-Resourced Speech Recognition: Livvi-Karelian Case |
title_sort |
deep models for low-resourced speech recognition: livvi-karelian case |
publisher |
Multidisciplinary Digital Publishing Institute |
publishDate |
2023 |
url |
https://doi.org/10.3390/math11183814 |
genre |
karelian |
genre_facet |
karelian |
op_source |
Mathematics Volume 11 Issue 18 Pages: 3814 |
op_relation |
Network Science https://dx.doi.org/10.3390/math11183814 |
op_rights |
https://creativecommons.org/licenses/by/4.0/ |
op_doi |
https://doi.org/10.3390/math11183814 |
container_title |
Mathematics |
container_volume |
11 |
container_issue |
18 |
container_start_page |
3814 |
_version_ |
1779316365877837824 |