Description
Date depot: 27 mars 2021
Titre: Deep learning for the generation of realistic music recordings based a symbolic musical representation
Directeur de thèse:
Axel ROEBEL (STMS)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Signal et communications
Resumé: To overcome the issue of annotating datasets, the present doctoral research project aims to develop innovative algorithms to generate realistic music mixes based on a symbolic musical representation, that is the MIDI format. This work will not only make it possible to synthesize instrument mixes of music pieces for creative applications, but also it will allow generating large musical datasets from symbolic musical representations for training DNN. Using the annotation derived from the MIDI scores, we will be able to learn music analysis models for tasks such as: recognition of key, mode, and chord progression, automatic transcription, tempo estimation, instrument recognition, down-beat detection.
Doctorant.e: Renault Lenny