The invention pertains to a method for predicting the attention of at least one audience during a presentation by at least one speaker. The method is such that it comprises the steps of measurements (E25) of vocal or gestural characteristics of the at least one speaker of the presentation in progress and/or of measurements of characteristics of content of the presentation in progress, measurement (E26) of at least one parameter of duration or of occurrence of the measured characteristics, consultation (E27) of a database comprising a correspondence between vocal or gestural speaker characteristics and/or presentation content characteristics, parameters of duration or of occurrence which relate to these characteristics and information relating to the evolution of the attention level for these characteristics and these parameters and recovery of the information relating to the evolution of the attention level corresponding to the measurements performed, of presentation (E28) to the at least one speaker of the presentation, of a prediction of attention level on the basis of the information recovered relating to the evolution of the attention level. The invention also pertains to a learning phase for obtaining the correspondences of the database, to a prediction device implementing the method described as well as to a learning device implementing the learning phase.L'invention se rapporte à un procédé de prédiction de l'attention d'au moins un auditoire lors d'une présentation par au moins un locuteur. Le procédé est tel qu'il comporte les étapes de mesures (E25) de caractéristiques vocales ou gestuelles du au moins un locuteur de la présentation en cours et/ou de mesures de caractéristiques de contenu de la présentation en cours, mesure (E26) d'au moins un paramètre de durée ou d'occurrence des caractéristiques mesurées, consultation (E27) d'une base de données comportant une correspondance entre des caractéristiques vocales ou gestuelles de locuteur et/ou des caractér