MAGIC DUST FOR CROSS-LINGUAL ADAPTATION OF MONOLINGUAL WAV2VEC-2.0 - Laboratoire Informatique de l'Université du Maine Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

MAGIC DUST FOR CROSS-LINGUAL ADAPTATION OF MONOLINGUAL WAV2VEC-2.0

Résumé

We propose a simple and effective cross-lingual transfer learning method to adapt monolingual wav2vec-2.0 models for Automatic Speech Recognition (ASR) in resource-scarce languages. We show that a monolingual wav2vec-2.0 is a good few-shot ASR learner in several languages. We improve its performance further via several iterations of Dropout Uncertainty-Driven Self-Training (DUST) by using a moderate-sized unlabeled speech dataset in the target language. A key finding of this work is that the adapted monolingual wav2vec-2.0 achieves similar performance as the topline multilingual XLSR model, which is trained on fifty-three languages, on the target language ASR task.
Fichier principal
Vignette du fichier
khurana22.pdf (123.82 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03544515 , version 1 (26-01-2022)

Identifiants

  • HAL Id : hal-03544515 , version 1

Citer

Sameer Khurana, Antoine Laurent, James Glass. MAGIC DUST FOR CROSS-LINGUAL ADAPTATION OF MONOLINGUAL WAV2VEC-2.0. ICASSP 2022, May 2022, Singapour, Singapore. ⟨hal-03544515⟩
151 Consultations
161 Téléchargements

Partager

Gmail Facebook X LinkedIn More