| Résumé |
In this paper, we present the various constituents of a spoken message which allow the observation of expressivity in speech. These constituents are joined into the perspective of the double coding of the speech, which distinguishes the linguistic channel of the paralinguistic channel in a spoken message. Among this last channel, several phenomena seem to participate in the demonstration of the expressivity: The prosody, naturally, but also the nonverbal sounds, as well as of possible restructurings. In a second part, we introduce the expressive French multi-speaker corpus: IrcamCorpusExpressivity. Several steps of labeling and analysis allow the examination of this corpus under the various angles corresponding to the constituents of the spoken message. These results can be used to improve the tasks of recognition, transformation and synthesis of the expressivity in the speech, and so contribute to the anthropomorphisation of the Human-machine interfaces. |
