This paper presents some pitch features for the characterization of intermediate vocal productions from the CNRS-Musée de l'Homme sound archives, in the context of the DIADEMS inter-disciplinary project gathering researchers from ethnomusicology and speech signal processing. Different categories —chanting, singing, recitation, storytelling, talking, lament— have been identified and characterized by ethnomusicologists and are confronted by acoustic analysis. A database totalizing 79 utterances from 25 countries spread around the world is used. Among the tested features , the note duration distribution has proved to be a relevant measure. Categories are mostly characterized by the proportion of 100-ms notes and the duration of the longest note. An evaluation of these features has been realized through a supervised classification using the different vocal categories. Classification results show that these two features allow a good discrimination between "speech", "chanting" and "singing", but are not suited for discriminating between the "speech" subcategories "recitation", "storytelling" and "talking".