Song: Come To Ge Ther
Team members: Sven Ahlbäck, Pietro Bolcato and Bob L.T. Sturm
On the internet: GitHub
This team from Sweden consists of professor Sven Ahlbäck, associate professor Bob L.T. Sturm, and master’s student Pietro Bolcato. They are currently attached to KMH (Royal Institute of Music), KTH (Royal Institute of Technology), and the company Doremir Music Research.
‘This competition is a perfect case study to test modern day technologies.’
Team KTH/KMH+Doremir went with a challenging approach. Instead of generating melody and lyrics separately from each other, they used one algorithm to create both. The team used AI to create one pitch of a certain length, and then based on that created a matching syllable. They repeated this process until they started generating sentences that made a reasonable amount of sense. As far as they know, they’re the first people to use this particular method.
Is team KTH/KMH+Doremir pleased with the result? ‘Building a neural network and training it is a milestone. But the real achievement lies in generating a dynamic melody and lyrics that make sense. This song, created by both humans and AI, felt more like a puzzle than a collaboration’. To Pietro Bolcato, the process felt like a creativity booster, because it provided some solid fundamental building blocks: the bassline and the lead melody as well as the lyrics. Every other element was built around that, from the harmony to the accompanying elements such as the drums. Speech synthesis provided by Google WaveNet was the cherry on top, while the AI generated melody and lyrics was sung by a young female singer, Olga Rosenberg.
The team members come from a variety of backgrounds in computer science, folk music and electrical and computer engineering. All three are active in the space where music and computers meet. Sven Ahlbäck leads the research and development department at Doremir Music Research, the company behind award-winning software ScoreCloud, which has figured out how to automate music notation. Noteworthy fact he is considered one of Sweden’s best fiddle players. Pietro Bolcato has experience with both emotional recognition and music suggestion by AI and is about to finish his second year at KTH. Bob L.T. Sturm hails from the United States and moved to Europe in 2009 to start his career in academic research. He’s also an associate professor in the Computer Sciences department at KTH and has previously worked on a machine learning algorithm that generated 100.000 new folk tunes. With two PhD’s and several MA’s on their resumes, this team has plenty of academic knowledge to throw at their competitors.
We Come to ge ther
As we dance
When we go get lost
That's how we go when
A-I will sing and a dance there will be
Our love, it feels like is growing in me