The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Os sistemas de reconhecimento automático de fala (ASR) multilíngue ponta a ponta (E2E) visam reconhecer falas multilíngues em uma estrutura unificada. Na atual estrutura ASR multilíngue E2E, a previsão de saída para um idioma específico não possui restrições no escopo de saída das unidades de modelagem. Neste artigo, uma estratégia de treinamento de supervisão de linguagem é proposta com máscaras de linguagem para restringir a distribuição de saída da rede neural. Para simular o cenário ASR multilíngue com informações de identidade de idioma desconhecidas, um classificador de identificação de idioma (LID) é aplicado para estimar as máscaras de idioma. Em quatro corpora de Babel, o sistema ASR multilíngue E2E proposto alcançou uma redução média da taxa absoluta de erros de palavras (WER) de 2.6% em comparação com o sistema de linha de base multilíngue.
Danyang LIU
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Ji XU
Chinese Academy of Sciences,University of Chinese Academy of Sciences
Pengyuan ZHANG
Chinese Academy of Sciences,University of Chinese Academy of Sciences
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Danyang LIU, Ji XU, Pengyuan ZHANG, "End-to-End Multilingual Speech Recognition System with Language Supervision Training" in IEICE TRANSACTIONS on Information,
vol. E103-D, no. 6, pp. 1427-1430, June 2020, doi: 10.1587/transinf.2019EDL8214.
Abstract: End-to-end (E2E) multilingual automatic speech recognition (ASR) systems aim to recognize multilingual speeches in a unified framework. In the current E2E multilingual ASR framework, the output prediction for a specific language lacks constraints on the output scope of modeling units. In this paper, a language supervision training strategy is proposed with language masks to constrain the neural network output distribution. To simulate the multilingual ASR scenario with unknown language identity information, a language identification (LID) classifier is applied to estimate the language masks. On four Babel corpora, the proposed E2E multilingual ASR system achieved an average absolute word error rate (WER) reduction of 2.6% compared with the multilingual baseline system.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2019EDL8214/_p
Copiar
@ARTICLE{e103-d_6_1427,
author={Danyang LIU, Ji XU, Pengyuan ZHANG, },
journal={IEICE TRANSACTIONS on Information},
title={End-to-End Multilingual Speech Recognition System with Language Supervision Training},
year={2020},
volume={E103-D},
number={6},
pages={1427-1430},
abstract={End-to-end (E2E) multilingual automatic speech recognition (ASR) systems aim to recognize multilingual speeches in a unified framework. In the current E2E multilingual ASR framework, the output prediction for a specific language lacks constraints on the output scope of modeling units. In this paper, a language supervision training strategy is proposed with language masks to constrain the neural network output distribution. To simulate the multilingual ASR scenario with unknown language identity information, a language identification (LID) classifier is applied to estimate the language masks. On four Babel corpora, the proposed E2E multilingual ASR system achieved an average absolute word error rate (WER) reduction of 2.6% compared with the multilingual baseline system.},
keywords={},
doi={10.1587/transinf.2019EDL8214},
ISSN={1745-1361},
month={June},}
Copiar
TY - JOUR
TI - End-to-End Multilingual Speech Recognition System with Language Supervision Training
T2 - IEICE TRANSACTIONS on Information
SP - 1427
EP - 1430
AU - Danyang LIU
AU - Ji XU
AU - Pengyuan ZHANG
PY - 2020
DO - 10.1587/transinf.2019EDL8214
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E103-D
IS - 6
JA - IEICE TRANSACTIONS on Information
Y1 - June 2020
AB - End-to-end (E2E) multilingual automatic speech recognition (ASR) systems aim to recognize multilingual speeches in a unified framework. In the current E2E multilingual ASR framework, the output prediction for a specific language lacks constraints on the output scope of modeling units. In this paper, a language supervision training strategy is proposed with language masks to constrain the neural network output distribution. To simulate the multilingual ASR scenario with unknown language identity information, a language identification (LID) classifier is applied to estimate the language masks. On four Babel corpora, the proposed E2E multilingual ASR system achieved an average absolute word error rate (WER) reduction of 2.6% compared with the multilingual baseline system.
ER -