Skip to Main content Skip to Navigation
Master thesis

Apprentissage par renforcement pour l'improvisation musicale automatique

Rémi Decelle 1
1 MULTISPEECH - Speech Modeling for Facilitating Oral-Based Communication
Inria Nancy - Grand Est, LORIA - NLPKD - Department of Natural Language Processing & Knowledge Discovery
Abstract : The DYCI2 ANR project aims to explore interactions between humans and artificial agents in the field of music improvisation. Interactive learning, one of the main's project research area, proposes models able to detect musical structures. All of these research are based on the OMax paradigm, an automatic music improvisation system. Unfortunately, such a system still creates false notes. Is it possible to have the system learned not to do those mistakes again ? To answer that question, we propose a melodic classification into two classes : the ones which has at least one wrong note, and the others which have no ones. This classification allows us to enhance the system by using an reinforcement learning algorithm. After introducing some musical words and explaining what are LSTMs, we present our neural network model which is going to classify melodies. We also propose a musical encoding scheme. We use Deep Q-Learning as reinforcement learning algorithm to improve the current system. We evaluate our neural network model with classical criteria. The final enhancement will be evaluate by listening to the melodies. At last, we discuss about our strategies.
Document type :
Master thesis
Complete list of metadata

Cited literature [35 references]  Display  Hide  Download
Contributor : Rémi Decelle Connect in order to contact the contributor
Submitted on : Thursday, September 21, 2017 - 2:53:00 PM
Last modification on : Saturday, October 16, 2021 - 11:26:09 AM


  • HAL Id : hal-01591521, version 1


Rémi Decelle. Apprentissage par renforcement pour l'improvisation musicale automatique. Intelligence artificielle [cs.AI]. 2017. ⟨hal-01591521⟩



Record views


Files downloads