LUX-ASR: Building an ASR system for the Luxembourgish language

[en] We present a first system for automatic speech recognition (ASR) for the low-resource language Luxembourgish. By applying transfer-learning, we were able to fine-tune Meta’s wav2vec2-xls-r-300m checkpoint with 35 hours of labeled Luxembourgish speech data. The best word error rate received lies at 14.47.

Disciplines :

Computer science

Author, co-author :

GILLES, Peter ; University of Luxembourg > Faculty of Humanities, Education and Social Sciences (FHSE) > Department of Humanities (DHUM)

HOSSEINI KIVANANI, Nina ; University of Luxembourg > Faculty of Science, Technology and Medicine (FSTM) > Department of Computer Science (DCS)

HILLAH, Léopold Edem Ayité ; University of Luxembourg > Faculty of Science, Technology and Medecine (FSTM)

External co-authors :

Language :

English

Title :

LUX-ASR: Building an ASR system for the Luxembourgish language

Publication date :

2023

Event name :

2022 IEEE Spoken Language Technology Workshop (SLT) SLT 2022

Event organizer :

SLT

Event place :

Doha, Qatar

Event date :

from 09-01-2023 to 12-01-2023

Audience :

International

Main work title :

Proceedings - 2022 IEEE Spoken Language Technology Workshop (SLT)

Editor :

IEEE, Spoken Language Technology

ISBN/EAN :

978-1-6654-7189-3

Pages :

1147-1149

Peer reviewed :

Peer reviewed

Focus Area :

Computational Sciences

Available on ORBilu :

since 09 May 2023

Statistics

Number of views

295 (17 by Unilu)

Number of downloads

220 (19 by Unilu)

More statistics