AI-Generated Music

A musical keyboard that incorporates the latest advancements in machine learning is designed to create original music from a simple input melody. Known as AWS DeepComposer, the system uses generative artificial intelligence (AI) – a technique that “connects the dots” between creative concepts and generates something that is totally new.

In generative AI, two networks (often called generative adversarial networks, or GANs) effectively work to train each other. The first is the generator, which takes the input and creates the output. The second is known as the discriminator. Its function is to provide feedback on whether the output is good or bad, as well as suggestions for improving it. This feedback loop takes only seconds in the SoundCloud, and the result is music that has never been heard before.

The melody is input via a 32-key, 2-octave keyboard, or a virtual keyboard on a touchscreen. The user can select from a variety of musical models, including rock, pop, jazz and classical, and the system will automatically generate a finished song, complete with accompaniments including guitar, bass, synthesizer and drums. In addition, a build-your-own model allows developers to select sample songs from a variety of genres to train the generative algorithm.

DeepComposer is the third in a family of machine learning devices to enhance human-AI collaboration, which also includes DeepRacer (a cloud-based racing simulator) and DeepLens (a camera for developing deep learning-based computer vision apps).