Skip to Main content Skip to Navigation

Résumé automatique de parole pour un accès efficace aux bases de données audio

Abstract : The digital era has revealed new ways to store great quantities of speech at a low cost. Whereas recent advances in spoken document retrieval, exploiting audio documents is still difficult because of the time necessary to listen to them. We try to attenuate this disadvantage by producing an automatic spoken abstract from the most important information. For that purpose, an extractive summarization algorithm is applied to the spoken content thanks to automatic speech structuring. The rich transcription is carried out thanks to Speeral and Alize toolkits developed at LIA. We complement this structuring chain by sentence segmentation and named entities detection, two important features for extractive summarization. The proposed summarization approach includes constraints imposed by audio data and interactions with the user. Moreover, the method integrates a projection of sentences in pseudo-semantic-space. We integrated the various modules in a coherent prototype that ease the study of user interactions. Due to the lack of evaluation data for the speech summarization task, we evaluate our approach on the textual documents from the DUC 2006 campaign. We simulate the impact of spoken content structuring by artificially degrading the textual content provided for DUC. Finally, the whole processing sequence is implemented within a demonstrator facilitating the access radio broadcasts from the ESTER evaluation campain. Within the framework of this prototype, we present an interactive timeline that aims at recontextualizing the spoken summary.
Document type :
Complete list of metadatas

Cited literature [62 references]  Display  Hide  Download
Contributor : Benoit Favre <>
Submitted on : Tuesday, January 5, 2010 - 4:47:40 PM
Last modification on : Tuesday, January 14, 2020 - 10:38:05 AM
Long-term archiving on: : Thursday, October 18, 2012 - 11:55:35 AM


  • HAL Id : tel-00444105, version 1



Benoit Favre. Résumé automatique de parole pour un accès efficace aux bases de données audio. Interface homme-machine [cs.HC]. Université d'Avignon, 2007. Français. ⟨tel-00444105⟩



Record views


Files downloads