Skip to main content

Flow Machines: Interacting with Style

Final Report Summary - FLOWMACHINES (Flow Machines: Interacting with Style)

The Flow-Machines project addressed the issue of building a new generation of tools for creators, using artificial intelligence to model the style of existing sets of works, in music and text. Based on the idea that creation often consists in extending existing styles, we developed a series of techniques to extract style from a set of works and generate new works in those styles. We obtained many strong results in this area, in particular about the automatic generation of stylistically consistent sequences from existing musical works, both in the symbolic domain (scores) and audio. Two scientific monographs describe these results in detail, as well as many scientific papers.
From a practical perspective, we developed a set of tools for creators, and worked with many musicians to study how to use these tools in musically meaningful and interesting ways. We produced the first ever pop music concert made with AI (October 2016), and the first AI-composed music album, including many international artists (Stromae, Kiesza, etc.). The success of the project led to a huge media coverage (Daddy's car, a song in the style of the Beatles got 1,7 million views on YouTube). Thanks to a PoC grant, we created the first AI-music label (Flow-Records), funded by Sony Europe, and promoted by Spotify (the main music distributor today). The research was finally relocated to Spotify and will double in size, showing the importance of the domain in the world of AI an entertainment, and the success of the project in the now very competitive world of music and artificial intelligence.