Skip to Main content Skip to Navigation
Master Thesis

Corpus oral pour les mesures de confiance pour la Traduction Automatique de la Parole

Abstract : The automatic speech translation combines two related domains, i.e. the automatic speech recognition and Machine translation. In order to improve the output of a translation system; and, hence, minimize the error rate, it's necessary to use confidence measures which can efficiently evaluate the accuracy of the hypothesis. However, there has been no corpus with such measures, so far. Therefore, to fill in this gap, we have first added oral recordings to an existing corpus and later, experimented with the whole of our new data. In this paper, we describe among other things strategy we adopted to classify the obtained labels such G: (Good) or B: (Bad). We obtained good results with good words (65,49%) and encouraging ones with words rated as bad (34.51%).
Document type :
Master Thesis
Complete list of metadata

Cited literature [11 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-01023835
Contributor : Uga - Bulles <>
Submitted on : Tuesday, July 15, 2014 - 12:04:50 PM
Last modification on : Tuesday, May 11, 2021 - 11:36:17 AM
Long-term archiving on: : Friday, November 21, 2014 - 5:21:36 PM

Identifiers

  • HAL Id : dumas-01023835, version 1

Citation

Marwa Hadj Salah. Corpus oral pour les mesures de confiance pour la Traduction Automatique de la Parole. Sciences de l'Homme et Société. 2014. ⟨dumas-01023835⟩

Share

Metrics

Record views

494

Files downloads

1671