Description
Date depot: 9 avril 2021
Titre: End-to-end Automated Drum Transcription
Directeur de thèse:
Florent JACQUEMARD (CEDRIC)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé: This thesis aims at studying the automation of the transcription of drummer's performances into common music notation. We shall in particular investigate the integration of state-of-the-art acoustic models for the extraction of low-level features (such as onset times of notes and instrument class) with language models describing precisely the nature of notation expected in output. Beyond the case study of drum transcription, it will also be interesting to study how using language models can improve the results of acoustic models, and reciprocally, a long-standing question Music Information Retrieval.