Utilisation des grammaires probabilistes dans les t\^aches de segmentation et d'annotation prosodique

TL;DR

This paper introduces a hybrid symbolic-probabilistic approach using minimal hierarchical probabilistic grammars for speech segmentation and prosodic annotation, with thorough qualitative and quantitative evaluation.

Contribution

It presents a novel method combining symbolic and probabilistic techniques with minimal hierarchical grammars for prosodic speech segmentation.

Findings

Probabilistic grammars effectively predict prosodic segmentation.

Qualitative and quantitative evaluations demonstrate the approach's robustness.

The minimal hierarchical structure simplifies grammar construction.

Abstract

Nous pr\'esentons dans cette contribution une approche \`a la fois symbolique et probabiliste permettant d'extraire l'information sur la segmentation du signal de parole \`a partir d'information prosodique. Nous utilisons pour ce faire des grammaires probabilistes poss\'edant une structure hi\'erarchique minimale. La phase de construction des grammaires ainsi que leur pouvoir de pr\'ediction sont \'evalu\'es qualitativement ainsi que quantitativement. ----- Methodologically oriented, the present work sketches an approach for prosodic information retrieval and speech segmentation, based on both symbolic and probabilistic information. We have recourse to probabilistic grammars, within which we implement a minimal hierarchical structure. Both the stages of probabilistic grammar building and its testing in prediction are explored and quantitatively and qualitatively evaluated.