Constitution d'un corpus de traduction de la parole : augmentation du corpus LibriSpeech

Abstract : Large quantities of parallel corpora such as Europarl, OpenSubitles, etc. is available for machine translation systems. However, for speech translation systems, number of available corpora is very limited. In this research paper, we investigate building a large scale (236h) non-synthetic corpus for speech translation systems from prepared speech recordings of LibriSpeech project. First, we gathered available French e-books corresponding to the English audio-books from LibriSpeech. Then, after aligning English and French texts at the sentence level, we also aligned speech segments at the sentence level with their respective translations. This allowed us to obtain an alignment at the utterance level aligned with their translations. Lastly, we manually evaluated 200 alignments and added alignment scores between transcriptions and their respective translations to sort the corpus according to those scores.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [23 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-01712400
Contributor : Uga - Bulles <>
Submitted on : Monday, February 19, 2018 - 2:37:21 PM
Last modification on : Tuesday, April 2, 2019 - 1:48:16 AM
Long-term archiving on : Monday, May 7, 2018 - 12:37:13 AM

File

KOCABIYIKOGLU Ali Can M2 IDL.p...
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-01712400, version 1

Citation

Ali Kocabıyıkoğlu. Constitution d'un corpus de traduction de la parole : augmentation du corpus LibriSpeech. Sciences de l'Homme et Société. 2017. ⟨dumas-01712400⟩

Share

Metrics

Record views

102

Files downloads

111