L'analyse automatique de la musique et les archives sonores du 'Musée de l'Homme'

International audience; The Musée de l’Homme has collected and preserved traditional music recorded all over the world from 1900 to the present day. This heritage contains mainly vocal music, with or without instruments. Spoken voice interventions are also found in field recordings. Today, this collection stands among the most important ones in Europe : about 53,000 sound files are available in a database on line (28,000 are on free access for the public): https://archives.crem-cnrs.fr/Managed by the French Research Center for Ethnomusicology (CREM), these Sound Archives database is accessible through an "open source" web platform. The collaborative nature of the platform enables users to continuously enrich the description of the audio or video documents. Researchers, teachers, students or musicians from various countries, can work together and conduct research from both synchronic and diachronic perspectives.Since 2013, this database offers the opportunity to develop computational analysis tools, in order to help the archiving process and to facilitate the search for musical information in the sound content: speech detection, singing voice detection, music detection, monody/polyphony detection. The sound visualisation is also helpful to the musical transcription. some tools are open access, others with a user accountIn this poster, we will show examples of innovative research programs generated by this sound database : the DIADEMS project (Description, Indexation, Access, Document EthnoMusicology and sound, a French Musical Information Retrieval research project), but also projects with the Queen Mary University and the New York University.The benefits of those collaborative works apply to numerous aspects of the field of ethnomusicology, ranging from musical analysis to comparative anthropology of music, as well as to the fields of linguistics and musical acoustic.