@inproceedings{gupta-boulianne-2022-progress,
title = "Progress in Multilingual Speech Recognition for Low Resource Languages {K}urmanji {K}urdish, {C}ree and Inuktut",
author = "Gupta, Vishwa and
Boulianne, Gilles",
editor = "Calzolari, Nicoletta and
B{\'e}chet, Fr{\'e}d{\'e}ric and
Blache, Philippe and
Choukri, Khalid and
Cieri, Christopher and
Declerck, Thierry and
Goggi, Sara and
Isahara, Hitoshi and
Maegaard, Bente and
Mariani, Joseph and
Mazo, H{\'e}l{\`e}ne and
Odijk, Jan and
Piperidis, Stelios",
booktitle = "Proceedings of the Thirteenth Language Resources and Evaluation Conference",
month = jun,
year = "2022",
address = "Marseille, France",
publisher = "European Language Resources Association",
url = "https://aclanthology.org/2022.lrec-1.689",
pages = "6420--6428",
abstract = "This contribution presents our efforts to develop the automatic speech recognition (ASR) systems for three low resource languages: Kurmanji Kurdish, Cree and Inuktut. As a first step, we generate multilingual models from acoustic training data from 12 different languages in the hybrid DNN/HMM framework. We explore different strategies for combining the phones from different languages: either keep the phone labels separate for each language or merge the common phones. For Kurmanji Kurdish and Inuktut, keeping the phones separate gives much lower word error rate (WER), while merging phones gives lower WER for Cree. These WER are lower than training the acoustic models separately for each language. We also compare two different DNN architectures: factored time delay neural network (TDNN-F), and bidirectional long short-term memory (BLSTM) acoustic models. The TDNN-F acoustic models give significantly lower WER for Kurmanji Kurdish and Cree, while BLSTM acoustic models give significantly lower WER for Inuktut. We also show that for each language, training multilingual acoustic models by one more epoch with acoustic data from that language reduces the WER significantly. We also added 512-dimensional embedding features from cross-lingual pre-trained wav2vec2.0 XLSR-53 models, but they lead to only a small reduction in WER.",
}
<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="gupta-boulianne-2022-progress">
<titleInfo>
<title>Progress in Multilingual Speech Recognition for Low Resource Languages Kurmanji Kurdish, Cree and Inuktut</title>
</titleInfo>
<name type="personal">
<namePart type="given">Vishwa</namePart>
<namePart type="family">Gupta</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Gilles</namePart>
<namePart type="family">Boulianne</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2022-06</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the Thirteenth Language Resources and Evaluation Conference</title>
</titleInfo>
<name type="personal">
<namePart type="given">Nicoletta</namePart>
<namePart type="family">Calzolari</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Frédéric</namePart>
<namePart type="family">Béchet</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Philippe</namePart>
<namePart type="family">Blache</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Khalid</namePart>
<namePart type="family">Choukri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Christopher</namePart>
<namePart type="family">Cieri</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Thierry</namePart>
<namePart type="family">Declerck</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sara</namePart>
<namePart type="family">Goggi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hitoshi</namePart>
<namePart type="family">Isahara</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bente</namePart>
<namePart type="family">Maegaard</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joseph</namePart>
<namePart type="family">Mariani</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hélène</namePart>
<namePart type="family">Mazo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jan</namePart>
<namePart type="family">Odijk</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Stelios</namePart>
<namePart type="family">Piperidis</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>European Language Resources Association</publisher>
<place>
<placeTerm type="text">Marseille, France</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
</relatedItem>
<abstract>This contribution presents our efforts to develop the automatic speech recognition (ASR) systems for three low resource languages: Kurmanji Kurdish, Cree and Inuktut. As a first step, we generate multilingual models from acoustic training data from 12 different languages in the hybrid DNN/HMM framework. We explore different strategies for combining the phones from different languages: either keep the phone labels separate for each language or merge the common phones. For Kurmanji Kurdish and Inuktut, keeping the phones separate gives much lower word error rate (WER), while merging phones gives lower WER for Cree. These WER are lower than training the acoustic models separately for each language. We also compare two different DNN architectures: factored time delay neural network (TDNN-F), and bidirectional long short-term memory (BLSTM) acoustic models. The TDNN-F acoustic models give significantly lower WER for Kurmanji Kurdish and Cree, while BLSTM acoustic models give significantly lower WER for Inuktut. We also show that for each language, training multilingual acoustic models by one more epoch with acoustic data from that language reduces the WER significantly. We also added 512-dimensional embedding features from cross-lingual pre-trained wav2vec2.0 XLSR-53 models, but they lead to only a small reduction in WER.</abstract>
<identifier type="citekey">gupta-boulianne-2022-progress</identifier>
<location>
<url>https://aclanthology.org/2022.lrec-1.689</url>
</location>
<part>
<date>2022-06</date>
<extent unit="page">
<start>6420</start>
<end>6428</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Progress in Multilingual Speech Recognition for Low Resource Languages Kurmanji Kurdish, Cree and Inuktut
%A Gupta, Vishwa
%A Boulianne, Gilles
%Y Calzolari, Nicoletta
%Y Béchet, Frédéric
%Y Blache, Philippe
%Y Choukri, Khalid
%Y Cieri, Christopher
%Y Declerck, Thierry
%Y Goggi, Sara
%Y Isahara, Hitoshi
%Y Maegaard, Bente
%Y Mariani, Joseph
%Y Mazo, Hélène
%Y Odijk, Jan
%Y Piperidis, Stelios
%S Proceedings of the Thirteenth Language Resources and Evaluation Conference
%D 2022
%8 June
%I European Language Resources Association
%C Marseille, France
%F gupta-boulianne-2022-progress
%X This contribution presents our efforts to develop the automatic speech recognition (ASR) systems for three low resource languages: Kurmanji Kurdish, Cree and Inuktut. As a first step, we generate multilingual models from acoustic training data from 12 different languages in the hybrid DNN/HMM framework. We explore different strategies for combining the phones from different languages: either keep the phone labels separate for each language or merge the common phones. For Kurmanji Kurdish and Inuktut, keeping the phones separate gives much lower word error rate (WER), while merging phones gives lower WER for Cree. These WER are lower than training the acoustic models separately for each language. We also compare two different DNN architectures: factored time delay neural network (TDNN-F), and bidirectional long short-term memory (BLSTM) acoustic models. The TDNN-F acoustic models give significantly lower WER for Kurmanji Kurdish and Cree, while BLSTM acoustic models give significantly lower WER for Inuktut. We also show that for each language, training multilingual acoustic models by one more epoch with acoustic data from that language reduces the WER significantly. We also added 512-dimensional embedding features from cross-lingual pre-trained wav2vec2.0 XLSR-53 models, but they lead to only a small reduction in WER.
%U https://aclanthology.org/2022.lrec-1.689
%P 6420-6428
Markdown (Informal)
[Progress in Multilingual Speech Recognition for Low Resource Languages Kurmanji Kurdish, Cree and Inuktut](https://aclanthology.org/2022.lrec-1.689) (Gupta & Boulianne, LREC 2022)
ACL