Emotional speech characterization for real time applications in real environments

A simple and effective method of automatic discrimination between emotional and unemotional speech is presented. Traditional methods of emotional discrimination use prosodic and paralinguistic features, which are determined by a linguistic segmentation of the locution. However, these methods are not appropriate in real time applications because of their high computational cost and the linguistic segmentation requirement by locutions. This letter proposes a new strategy based on a few prosodic and paralinguistic features set obtained from a temporal segmentation of the speech signal. This new strategy is robust to interfering noises that are present in real environments, offering a low computational cost and improving the performance of a segmentation based on linguistic aspects.