Amazon Web Services debuted a keyboard called DeepComposer this week, claiming it’s “the world’s first musical keyboard powered by generative AI.” It has 32 keys, costs $99, and connects to a software interface that uses machine learning and cloud computing to generate music based on what you play.
It’s been unclear who this is for, and many have latched on to the fact that the music it creates just sounds bad. It looks like a consumer product, and Amazon used an over-the-top presentation to hype it, which included what AWS claimed was “the first hybrid AI human pop acoustic collaboration.” (It’s not.)
But actually, the keyboard is intended to be a beginning tool for developers to get into machine learning and music. The device is AWS’s newest offering for developers to familiarize themselves with aspects of machine learning, following AWS DeepRacer (an RC car) and AWS DeepLens (a camera).
DeepComposer is not meant to make music for entertainment purposes or push the state of generative AI. It will never be marketed to aspiring musicians.
I took a deep look at AWS #DeepComposer and I have to say that I have no idea what to do with this nor who it is for. I can’t imagine training ML models by playing thousands of tracks on a physical keyboard. Then what would I do with the inferred MIDI? How is that teaching me ML?
Even though Amazon says DeepComposer is for developers, many devs don’t understand what to do with it. According to Amazon, since this is for developers, “no musical knowledge” is needed. But it still uses traditional music theory terms and is centered on a thing that very much requires a modicum of musical knowledge — a keyboard.
The $99 physical keyboard isn’t even necessary since the DeepComposer software has a virtual keyboard. And AWS didn’t develop or design the keyboard. It’s a MIDI controller by Taiwanese company Midiplus that’s been around for years. Amazon sells it for $46.15. They’re the same. Amazon’s costlier version is not “powered by AI,” it sends MIDI to software that’s hooked up to the cloud. Any MIDI keyboard would technically work.
At the end of the day, this is not for the everyday person. AWS does not claim it is. DeepComposer is not supposed to write the next radio hit. But it’s also hard to see how it’s meaningful for its target audience.
It is compelling to see a company as large as Amazon getting into the world of democratizing music creation, even if this first iteration somewhat missed the mark. Who knows, maybe it’s all a long play to get more devs familiar with AWS’s machine learning platform, SageMaker.
But given how simple DeepComposer is, developers could learn about the same basics of machine learning and music through any number of other interfaces, like Google Magenta, without being pitched an overpriced controller stamped with a logo.