Supervision : Amal EL FALLAH SEGHROUCHNI

Co-supervision : BRIOT Jean-Pierre

Deep Learning for Conditioned Multi-track Symbolic Music Generation

We study in this thesis the application of artificial intelligence, especially deep learning, to music generation. The main interest is to put this discipline at the service of musicians in order to help them in their creative process, by exploiting their ideas to transcribe, extend or modify them according to their desires. A first axis consists in the creation of algorithmic and mathematical models and methods pushing the limits of existing solutions in the literature. The second axis focuses on the adaptation and encoding of musical data into representations responding to the constraints of the associated models and methods while offering as many controls as possible in the creative process that will influence the results according to the user's wishes.

2021-2023 Publications

• 2023
  • N. Fradet, N. Gutowski, F. Chhel, J.‑P. Briot : “Impact of time and note duration tokenizations on deep learning symbolic music modeling”, 24^th Conference of the International Society for Music Information Retrieval (ISMIR) 2023, Milano, Italy (2023)
  • N. Fradet, J.‑P. Briot, F. Chhel, A. El Fallah‑Seghrouchni, N. Gutowski : “Byte Pair Encoding for Symbolic Music”, (2023)
• 2021
  • N. Fradet, J.‑P. Briot, F. Chhel, A. El Fallah‑Seghrouchni, N. Gutowski : “MidiTok: A Python Package for MIDI File Tokenization”, Extended Abstracts for the Late-Breaking Demo Session of the 22^nd International Society for Music Information Retrieval Conference, Online, United States (2021)