Germany - Ligatur

Team Ligatur consists of PhD student Ullika Scholz and data scientist André Röhrig. Both are fascinated by mathematical systems, especially where they fail. They are familiar with AI’s successes in, for example, image generation, but curious about whether complete automation can produce a high-quality piece of music. They confess it is also a (golden) opportunity to finally release their very own AI-song.

Ullika has been fascinated by programming since the first computer appeared in her house. This inspired her to choose computer science during high school, and eventually studying mathematics. André works in the financial world, where he uses deep learning models to detect fraud.

The duo see themselves as AI curators. In making their AI Song Contest entry, they kept human intervention to a minimum in a wonderful example of what this technology can and cannot achieve. They used AI to create the segments of a typical pop song, such as the verses, chorus, and bridge, and manually repeated these with audio software to generate a coherent song. The generated melodies were compared to the audio samples of Eurovision songs. They concluded that piano and strings are the most suitable to use as instruments. Generating the lyrics was a piece of cake. They first let the AI generate a hundred lyrics and then picked ten favorites manually. After generating the music, André and Ullika chose one that fitted the gloomy atmosphere of the audio the most.

As curators, the pair wanted to influence the piece as little as possible. But there is still humanity in it: André sang the lyrics, and his voice was brought to the same pitch as the midi with a vocoder. Then all the parts were put together without further post-processing, and voilà!