The program works by identifying the 12 standard musical notes in a sung melody, and then feeding those notes into an algorithm that has been trained by listening to 300 songs in varying genres and learning how to identify chords and melody fragments that work well together. The result is a series of musical accompaniments that users can adjust via sliders for "happy factor" and "jazz factor."
"I suspect musicians will argue that this is another step towards homogenized elevator music for all," Peter Bentley, a computer scientist at University College London, told New Scientist. "But I see a big market for this, whether it's liked by musicians or not." We agree, and we think the web is the perfect place to find that market.