Robohub.org
 

‘Earworm melodies with strange aspects’ – what happens when AI makes music


by
12 January 2018



share this:

A new AI machine creates new music from songs it’s fed, mimicking their style. Image credit – FlowMachines


by Kevin Casey

The first full-length mainstream music album co-written with the help of artificial intelligence (AI) was released on 12 January and experts believe that the science behind it could lead to a whole new style of music composition.

Popular music has always been fertile ground for technological innovation. From the electric guitar to the studio desk, laptops and the wah-wah pedal, music has the ability to absorb new inventions with ease.

Now, the release of Hello World, the first entire studio album co-created by artists and AI could mark a watershed in music composition.

Stemming from the FlowMachines project, funded by the EU’s European Research Council, the album is the fruits of the labour of 15 artists, music producer Benoit Carré, aka Skygge, and creative software designed by computer scientist and AI expert François Pachet.

Already Belgian pop sensation Stromae and chart-topping Canadian chanteuse Kiesza have been making waves with the single Hello Shadow.

The single Hello Shadow, featuring Stromae and Kiesza, is taken from the AI-co-written album, Hello World. Video credit – SKYGGE MUSIC

The software works by using neural networks – artificial intelligence systems that learn from experience by forming connections over time, thereby mimicking the biological networks of people’s brains. Pachet describes its basic job as ‘to infer the style of a corpus (of music) and generate new things’.

A musician firstly provides ‘inspiration’ to the software by exposing it to a collection of songs. Once the system understands the style required it outputs a new composition.

‘The system … analyses the music in terms of beats, melody and harmony,’ said Pachet, ‘And then outputs an original piece of music based on that style.’

Creative workflow

The design challenge with this software was to make it adapt to the creative workflow of musicians without becoming a nuisance.

‘The core (problem) was how to do that so that (it) takes into account user constraints. Why? Because if you compose music, actually you never do something from scratch from A to Z,’ said Pachet.

He outlines a typical scenario where the AI software generates something and only parts of it are useful but the musician wants to keep it in, drop the rest and generate new sounds using the previous partial output. It’s a complex requirement, in other words.

‘Basically, the main contribution of the project was to find ways to do that, to do that well and to do that fast,’ said Pachet. ‘It was really an algorithmic problem.’ As creative workers driven by intuition, musicians need direct results to maintain their momentum. A clunky tool with ambivalent results would not last long in a creative workflow.

Pachet is satisfied that his technical goal is completed and that the AI will generate music ‘quickly and under user constraints’.

After years of development and refinement, the AI music tool now fits on a laptop, such as to be found in any recording studio, anywhere. In the hands of music producer Carré, the application became the creative tool that built Hello World.

Computer scientist and AI expert Francois Pachet created a system that co-writes music. Image credit – Kevin Casey/ Horizon

Collaboration

As a record producer, Carré collaborated closely with the artists in the studio to write and produce songs. So, as the resident musical expert, can Carré say if this is a new form of music?

‘It’s not a new form of music,’ he said, ‘It’s a new way to create music.’

Carré said he believes the software could lead to a new era in composition. ‘Every time there is a new tool there is a new kind of compositional style. For this project we can see that there is a new kind of melody that was created.’ He describes this as ‘earworm melodies with strange aspects’.

He also says that the process is a real collaboration between human and machine. The system creates original compositions that are then layered into songs in various forms, whether as a beat, a melody or an orchestration. During the process, artists such as Stromae are actively involved in making decisions about what and how to include the muscial fragments the AI provides.

‘You can recognise all the artists because they have made choices that are their identity, I think,’ said Carré.

Pachet concurs. ‘You know in English you say every Lennon needs a McCartney – so that’s the kind of stuff we are aiming at. We are not aiming at autonomous creation. I don’t believe that’s interesting, I don’t believe it’s possible actually, because we have no clue how to give a computer a sense of agency, a sense that something is going somewhere, (that) it has some meaning, a soul, if you want.’

The album’s title Hello World reflects the expression commonly used the very first time someone runs a new computer program or starts a website as proof that is working. Carré believes that Hello World is just the first step and the software signals the start of a whole new way of composing.

‘Maybe not next year, but in five years there will be a new set of tools that helps creators to make music,’ said Carré.


More info

FlowMachines

Max Order web comic




Horizon Magazine brings you the latest news and features about thought-provoking science and innovative research projects funded by the EU.
Horizon Magazine brings you the latest news and features about thought-provoking science and innovative research projects funded by the EU.





Related posts :



RoboCup Logistics League: an interview with Alexander Ferrein, Till Hofmann and Wataru Uemura

and   25 Sep 2025
Find out more about the RoboCup league focused on production logistics and the planning.

Drones and Droids: a co-operative strategy game

  22 Sep 2025
Scottish Association for Marine Science is running a crowdfunding campaign for educational card game.

Call for AAAI educational AI videos

  22 Sep 2025
Submit your contributions by 30 November 2025.

Self-supervised learning for soccer ball detection and beyond: interview with winners of the RoboCup 2025 best paper award

  19 Sep 2025
Method for improving ball detection can also be applied in other fields, such as precision farming.

#ICML2025 outstanding position paper: Interview with Jaeho Kim on addressing the problems with conference reviewing

  15 Sep 2025
Jaeho argues that the AI conference peer review crisis demands author feedback and reviewer rewards.

Apertus: a fully open, transparent, multilingual language model

  11 Sep 2025
EPFL, ETH Zurich and the Swiss National Supercomputing Centre (CSCS) released Apertus today, Switzerland’s first large-scale, open, multilingual language model.

Robots to the rescue: miniature robots offer new hope for search and rescue operations

  09 Sep 2025
Small two-wheeled robots, equipped with high-tech sensors, will help to find survivors faster in the aftermath of disasters.

#IJCAI2025 distinguished paper: Combining MORL with restraining bolts to learn normative behaviour

and   04 Sep 2025
The authors introduce a framework for guiding reinforcement learning agents to comply with social, legal, and ethical norms.



 

Robohub is supported by:




Would you like to learn how to tell impactful stories about your robot or AI system?


scicomm
training the next generation of science communicators in robotics & AI


 












©2025.05 - Association for the Understanding of Artificial Intelligence