Model-based compensation of the additive noise for continuous speech recognition. experiments using the Aurora II database and tasks | Publicación