How we used our Music Engine to create the first AI-generated album of Chinese Music
Building AIVA, an Artificial Intelligence capable of composing emotional soundtracks, requires that a lot of our Research & Development revolves around making the soundtracks composed more personalised, so that composers can more easily work with our AI. AIVA was able to compose mostly cinematic scores — which is the predominant form of soundtrack music — up until now.
AI-composed World Music
Recently, we’ve also expanded to other styles. In movies and games, world music is often used to set different moods for different geographies. AIVA can now compose a wide range of world music (Tango, Middle Eastern, Chinese, etc.). And today, we are releasing on all major streaming platforms an entire album of Chinese music, composed by AI. We call it 艾娲 (ài wā).
We achieved this by using a combination of genetic algorithms and a new module of deep neural networks used to create musical variations.
Composing with Genetic Algorithms
Genetic algorithms are inspired by the principles of evolution: given a population of candidate solutions (in our case, musical scores), we make their properties evolve through processes akin to genetic mutation and chromosome crossovers to create new properties, which will be tested against some criteria of evolution. In music, those criteria are the rules of the style of music that we want the AI to compose in. Finally, the “fitter” scores are selected according to those criteria, and we can repeat this process until the evolutionary system produces pleasing results.
Much like film music has a single theme reprised several times throughout the soundtrack, we wanted to apply this idea of theme variation to our Chinese Album, and have some repetitions of themes present in different tracks. To achieve this, we applied techniques of deep learning to create musical variations of a theme in the Chinese album presented above. For this, we give a musical theme in MIDI format to our variation module, that then iteratively applies changes to the harmony, melody and rhythm of the piece, while staying faithful to the original theme, but moving away further from it with each iteration.
A Music Engine to supercharge composers
As a composer, writer’s block is probably one of the most dreadful part of the creation process. Especially when your day to day job involves the responsibility of composing hours long, high quality soundtracks in a matter of a couple of months at best, down to a couple of weeks at worst. Add to that the ever changing musical requirements in a fast paced production environment and the challenges of communication between composer and director, and compromises start to become a necessary part of a composer’s life.
We are currently building a Music Engine powered by AIVA, to supercharge composers’ creative abilities by leveraging AI to provide them with a lot of musical ideas to build from. Ultimately, we believe that writer’s block should be a thing of the past, so that composers can have more time to push their work — and themselves — towards excellence.
You can watch how Brad Frey, our Music Supervisor at AIVA, used the Music Engine to create an entire album of Chinese Music:
AIVA’s Music Engine is scheduled for beta release at the end of this year. You can sign up to the upcoming beta here.
Thanks for reading!
Would you use the Music Engine? Comment below
♢ JOIN THE COMMUNITY ♢
Special thanks to the whole team of people at AIVA who push the boundaries of AI & Music every day (Denis, Chien-Hung, Brad, Vincent, Ashkhen, Sebastian, Theo, Olivier, Aleksei, Michael — you guys rock!) and to all composers out there whose job it is to make us dream.