Fine-tuning pre-trained models for Automatic Speech Recognition: experiments on a fieldwork corpus of Japhug (Trans-Himalayan family)
Séverine Guillaume
(1)
,
Guillaume Wisniewski
(2)
,
Cécile Macaire
(1, 3)
,
Guillaume Jacques
(4)
,
Alexis Michaud
(1)
,
Benjamin Galliot
(1)
,
Maximin Coavoux
(3)
,
Solange Rossato
(3)
,
Minh-Châu Nguyễn
(3)
,
Maxime Fily
(1, 5)
Séverine Guillaume
- Fonction : Auteur
- PersonId : 12704
- IdHAL : severine-guillaume
- ORCID : 0000-0003-1772-2600
Guillaume Wisniewski
- Fonction : Auteur
- PersonId : 748468
- IdHAL : guillaume-wisniewski
- ORCID : 0000-0002-4445-080X
- IdRef : 128062290
Cécile Macaire
- Fonction : Auteur
- PersonId : 1120002
- IdHAL : cecile-macaire
- ORCID : 0000-0003-1407-8880
Guillaume Jacques
- Fonction : Auteur
- PersonId : 6342
- IdHAL : guillaume-jacques
- ORCID : 0000-0003-1220-2323
- IdRef : 087255545
Alexis Michaud
- Fonction : Auteur
- PersonId : 419
- IdHAL : alexis-michaud
- ORCID : 0000-0003-1165-2680
- IdRef : 095131507
Benjamin Galliot
- Fonction : Auteur
- PersonId : 1184597
- IdHAL : benjamin-galliot
Maximin Coavoux
- Fonction : Auteur
- PersonId : 13643
- IdHAL : maximin-coavoux
- ORCID : 0000-0003-4089-4558
Minh-Châu Nguyễn
- Fonction : Auteur
- PersonId : 173973
- IdHAL : minhchaunguyen
- ORCID : 0000-0001-6771-9990
- IdRef : 26043082X
Maxime Fily
- Fonction : Auteur
- PersonId : 754685
- IdHAL : maxime-fily
- ORCID : 0009-0006-8292-5358
- IdRef : 270285113
Résumé
This is a report on results obtained in the development of speech recognition tools intended to support linguistic documentation efforts. The test case is an extensive fieldwork corpus of Japhug, an endangered language of the Trans-Himalayan (Sino-Tibetan) family. The goal is to reduce the transcription workload of field linguists. The method used is a deep learning approach based on the language-specific tuning of a generic pre-trained representation model, XLS-R, using a Transformer architecture. We note difficulties in implementation, in terms of learning stability. But this approach brings significant improvements nonetheless. The quality of phonemic transcription is improved over earlier experiments; and most significantly, the new approach allows for reaching the stage of automatic word recognition. Subjective evaluation of the tool by the author of the training data confirms the usefulness of this approach.
Domaines
LinguistiqueFormat du dépôt | Fichier |
---|---|
Type de dépôt | Communication dans un congrès |
Titre |
en
Fine-tuning pre-trained models for Automatic Speech Recognition: experiments on a fieldwork corpus of Japhug (Trans-Himalayan family)
|
Résumé |
en
This is a report on results obtained in the development of speech recognition tools intended to support linguistic documentation efforts. The test case is an extensive fieldwork corpus of Japhug, an endangered language of the Trans-Himalayan (Sino-Tibetan) family. The goal is to reduce the transcription workload of field linguists. The method used is a deep learning approach based on the language-specific tuning of a generic pre-trained representation model, XLS-R, using a Transformer architecture. We note difficulties in implementation, in terms of learning stability. But this approach brings significant improvements nonetheless. The quality of phonemic transcription is improved over earlier experiments; and most significantly, the new approach allows for reaching the stage of automatic word recognition. Subjective evaluation of the tool by the author of the training data confirms the usefulness of this approach.
|
Auteur(s) |
Séverine Guillaume
1
, Guillaume Wisniewski
2
, Cécile Macaire
1, 3
, Guillaume Jacques
4
, Alexis Michaud
1
, Benjamin Galliot
1
, Maximin Coavoux
3
, Solange Rossato
3
, Minh-Châu Nguyễn
3
, Maxime Fily
1, 5
1
LACITO -
Langues et civilisations à tradition orale
( 406905 )
- 7, rue Guy Môquet, 94800, VILLEJUIF
- France
2
LLF - UMR7110 -
Laboratoire de Linguistique Formelle
( 1004855 )
- 8 Rue Albert Einstein, 75013 Paris
- France
3
LIG -
Laboratoire d'Informatique de Grenoble
( 1043301 )
- UMR 5217 - Laboratoire LIG - Bâtiment IMAG - 700 avenue Centrale - Domaine Universitaire de Saint-Martin-d’Hères
Adresse postale : CS 40700 - 38058 Grenoble cedex 9
Tél. : 04 57 42 14 00
- France
4
CRLAO -
Centre de Recherches Linguistiques sur l'Asie Orientale
( 1324 )
- 105, boulevard Raspail, 75006, PARIS
- France
5
GIPSA-SYLDO -
GIPSA - SYstèmes Linguistiques, Dialectologie et Oralité
( 1043341 )
- GIPSA-lab, 11 rue des Mathématiques, Grenoble Campus BP46, F-38402 SAINT MARTIN D'HERES CEDEX
- France
|
Langue du document |
Anglais
|
Date de production/écriture |
2022
|
Licence |
Paternité - Pas d'utilisation commerciale - Partage selon les Conditions Initiales
|
Actes |
Oui
|
Titre du congrès |
ComputEL-5 5th Workshop on Computational Methods for Endangered Languages (ComputEL-5)
|
Vulgarisation |
Non
|
Comité de lecture |
Oui
|
Source |
Proceedings ComputEL-5
|
Invité |
Non
|
Audience |
Internationale
|
Titre de la collection |
Proceedings of ComputEL-5: Fifth Workshop on the Use of Computational Methods in the Study of Endangered Languages
|
Date début congrès |
2022-05-26
|
Date fin congrès |
2022-05-27
|
Ville |
Dublin
|
Pays |
Irlande
|
URL du congrès ou éditeur |
https://computel-workshop.org/computel-5/
|
Domaine(s) |
|
Projet(s) ANR |
|
Mots-clés |
en
Automatic Speech Recognition
|
DOI | 10.18653/v1/2022.computel-1.21 |
Origine :
Fichiers produits par l'(les) auteur(s)
Loading...