--by Matt Scott
Today I wrote a killer drum beat using a hardcover book, a seal, a car engine, and a dollar bill sliding between fingers. The Infinite Drum Machine is an odd project. At first glance it looks like a silly novelty -- just a rudimentary drum machine that’d be familiar to anyone who’s messed with sequencing software before, albeit with a deep library of real world sounds to drive it. But the technology behind it makes it fascinating
Google's A.I. Experiment group has been conducting ongoing artificial intelligence experiments that deal with machine learning (another recent app that resulted from it is Quick, Draw!, which attempts to recognize users’ drawings) . Here, the Google folks fed thousands of everyday sounds to a computer but with no information on what each sound was. Then, using a machine learning-enabled algorithm called t-SNE (“t-distributed stochastic neighbor embedding”), the computer catalogued each noise and grouped them together by category.
Finally, the machine created map of all the sounds rendered as a multi-colored 2D image, placing similar sounds close together. So, for example, rattling sounds have their own cluster, ringing sounds are grouped among one another, and so on. And the computer made all these connections.
Which brings us back to the drum machine. At random, the simple web app will select four sounds from the database and instantly write a beat for you. The user can then edit the beat sequence to their heart’s content. Forget about more cowbell - how about more cow? Or... more cow’s milk pouring into a glass? Of course, some sound combination are more successful than others: the app gave me a great beat from a coffee cup being placed on a distant counter, car keys, a projector screen, and an electronic phone beep. But a combo of a jeans zipper, headphones, water pouring over concrete, and a magnet - not so much.
While the drum machine app is apparently just a fun way for the Google team to demonstrate their cataloguing experiment, it’s exciting to think what creativity may spring from it. Recording artists have been using found sounds in music for decades to be sure—especially for percussion—but having such a vast, readymade library like this could inspire musicians in a new way.