Détection de transcriptions incorrectes de parole non-native dans le cadre de l’apprentissage de langues étrangères (Detection of incorrect transcriptions of non-native speech in the context of foreign language learning) [in French]
1820Authors
nan
Abstract
Detection of incorrect transcriptions of non-native speech in the context of foreign lan- guage learning This article analyses the detection of incorrect transcriptions of non-native speech in the context of foreign language learning. The purpose is to detect and reject incorrect transcriptions (i.e. those for which the text does not correspond to the associated speech signal) while being tolerant to the pronunciation defects of non-native speech. The proposed approach exploits the comparison between two alignements : one constrained by the transcript which is being checked, with an other one unconstrained, corresponding to a phonetic decoding. Several criteria are described and combined via a logistic regression function. The article analyzes the influence of different settings, such as the impact of non-native pronunciation variants, the use of decision functions dependent on the length of the transcriptions, and the impact of learning decision functions on native or non-native speech. The performance evaluations are conducted both on native speech and non-native speech. MOTS-CLÉS : Apprentissage d’une langue étrangère, entrées incorrectes, parole non-native, variantes de prononciation, alignements contraint et non-contraint. KEYWORDS: Foreign language learning, incorrect transcriptions, non-native speech, pronuncia- tion variants, constrained and unconstrained alignements. 409 1
Keywords
Full Content
Actes de la conférence conjointe JEP-TALN-RECITAL 2012, volume 1: JEP, pages 409–416, Grenoble, 4 au 8 juin 2012. c ⃝2012 ATALA & AFCP Détection de transcriptions incorrectes de parole non-native dans le cadre de l’apprentissage de langues étrangères Luiza Orosanu Denis Jouvet Dominique Fohr Irina Illina Anne Bonneau INRIA - LORIA, 615 rue de Jardin Botanique 54600 Villers-les-Nancy {luiza.orosanu, denis.jouvet, dominique.fohr, irina.illina, anne.bonneau}@loria.fr RÉSUMÉ Cet article analyse la détection de transcriptions incorrectes de parole non-native dans le contexte de l’apprentissage de langues étrangères. L’objectif est de détecter et rejeter les transcriptions incorrectes (i.e. celles pour lesquelles le texte ne correspond pas au signal de parole associé) tout en étant tolérant aux défauts de prononciation inhérents à la parole non-native. L’approche proposée exploite la comparaison d’un alignement contraint par la transcription à vérifier avec un alignement non contraint correspondant à un décodage phonétique. Plusieurs critères de comparaison sont décrits et combinés par l’intermédiaire d’une fonction de régression logistique. L’article analyse l’influence de divers paramétrages comme l’impact des variantes de prononciation non-natives, l’utilisation de fonctions de décision spécifiques à la longueur des transcriptions, et l’impact d’un apprentissage de la fonction de décision avec la parole native ou non-native. Les évaluations de performances sont menées à la fois sur des corpus de parole natives et non-natives. ABSTRACT Detection of incorrect transcriptions of non-native speech in the context of foreign lan- guage learning This article analyses the detection of incorrect transcriptions of non-native speech in the context of foreign language learning. The purpose is to detect and reject incorrect transcriptions (i.e. those for which the text does not correspond to the associated speech signal) while being tolerant to the pronunciation defects of non-native speech. The proposed approach exploits the comparison between two alignements : one constrained by the transcript which is being checked, with an other one unconstrained, corresponding to a phonetic decoding. Several criteria are described and combined via a logistic regression function. The article analyzes the influence of different settings, such as the impact of non-native pronunciation variants, the use of decision functions dependent on the length of the transcriptions, and the impact of learning decision functions on native or non-native speech. The performance evaluations are conducted both on native speech and non-native speech. MOTS-CLÉS : Apprentissage d’une langue étrangère, entrées incorrectes, parole non-native, variantes de prononciation, alignements contraint et non-contraint. KEYWORDS: Foreign language learning, incorrect transcriptions, non-native speech, pronuncia- tion variants, constrained and unconstrained alignements. 409 1 Introduction L’aide à l’apprentissage des langues étrangères est un domaine d’application de la reconnaissance automatique de la parole qui s’est développé ces dernières années. L’objectif est de détecter et signaler à l’apprenant ses erreurs ou défauts de prononciation, afin qu’il puisse les corriger, et peu à peu améliorer sa maîtrise de la langue étrangère. L’une des principales difficultés pour ces systèmes est la détection et la localisation automatique des défauts de prononciation (Herron et al., 1999) tout en restant robuste à la parole non-native. Des méthodes ont été proposées pour déterminer un score de …
Content truncated to 500 words
Labeled Text
Detection of incorrect transcriptions of non-native speech in the context of foreign language learning This article analyses the detection of incorrect transcriptions of non-native speech in the context of foreign language learning. The purpose is to detect and reject incorrect transcriptions (i.e. those for which the text does not correspond to the associated speech signal) while being tolerant to the pronunciation defects of non-native speech. The proposed approach exploits the comparison between two alignements : one constrained by the transcript which is being checked, with an other one unconstrained, corresponding to a phonetic decoding. Several criteria are described and combined via a logistic regression function. The article analyzes the influence of different settings, such as the impact of non-native pronunciation variants, the use of decision functions dependent on the length of the transcriptions, and the impact of learning decision functions on native or non-native speech. The performance evaluations are conducted both on native speech and non-native speech. MOTS-CLÉS : Apprentissage d’une langue étrangère, entrées incorrectes, parole non-native, variantes de prononciation, alignements contraint et non-contraint. KEYWORDS: Foreign language learning, incorrect transcriptions, non-native speech, pronunciation variants, constrained and unconstrained alignements. 409 1
Publication Details
- Paper ID:
- 1820
- Year:
- 2012
- Publication:
- TIPSTER
- Publication Type:
- conf
- DOI:
- nan