Over the last few years, computers and algorithms have been increasingly used for several tasks, including art creation. Driven by a strong interest in music, my objective at Sony CSL is to develop innovative machine learning techniques to design new tools and software to stimulate artists’ creativity and helping them composing original music. Nowadays, through the use of synthesisers, DAWs or VSTs, the music creation process is more and more digitalised. However, most tools interact with the timbre of the sound while letting artists arrange notes manually. Therefore the purpose of my PhD is to include the machine already in the writing process, by suggesting original but relevant melodies or arrangements, based on what the artist already composed. My objective is therefore twofold: on the one hand, the tools I implement must use the most advanced deep learning techniques in the research field of music generation to produce convincing melodies, but on the other hand they must remain lightweight and easy to use by people that do not have technical skills in computer science. Dealing with these two constraints is vital for music composers to be able and willing to use these tools, and thus see the machine as an ally and not a competitor with regard to their art creation process.
