Hear What Music Sounds Like When It’s Created by Synthesizers Made with Artificial Intelligence

When syn­the­siz­ers like the Yama­ha DX7 became con­sumer prod­ucts, the pos­si­bil­i­ties of music changed for­ev­er, mak­ing avail­able a wealth of new, often total­ly unfa­mil­iar sounds even to musi­cians who’d nev­er before had a rea­son to think past the elec­tric gui­tar. But if the peo­ple at Project Magen­ta keep doing what they’re doing, they could soon bring about a wave of even more rev­o­lu­tion­ary music-mak­ing devices. That “team of Google researchers who are teach­ing machines to cre­ate not only their own music but also to make so many oth­er forms of art,” writes the New York Times’ Cade Metz, work toward not just the day “when a machine can instant­ly build a new Bea­t­les song,” but the devel­op­ment of tools that allow artists “to cre­ate in entire­ly new ways.”

Using neur­al net­works, “com­plex math­e­mat­i­cal sys­tems allow machines to learn spe­cif­ic behav­ior by ana­lyz­ing vast amounts of data” (the kind that gen­er­at­ed all those dis­turb­ing “Deep­Dream” images a while back), Magen­ta’s researchers “are cross­breed­ing sounds from very dif­fer­ent instru­ments — say, a bas­soon and a clavi­chord — cre­at­ing instru­ments capa­ble of pro­duc­ing sounds no one has ever heard.”

You can give one of the results of these exper­i­ments a test dri­ve your­self with NSynth, described by its cre­ators as “a research project that trained a neur­al net­work on over 300,000 instru­ment sounds.” Think of Nsynth as a syn­the­siz­er pow­ered by AI.

Fire it up, and you can mash up and play your own son­ic hybrids of gui­tar and sitar, pic­co­lo and pan flute, ham­mer dul­cimer and dog. In the video at the top of the post you can hear “the first tan­gi­ble prod­uct of Google’s Magen­ta pro­gram,” a short melody cre­at­ed by an arti­fi­cial intel­li­gence sys­tem designed to cre­ate music based on infer­ences drawn from all the music it has “heard.” Below that, we have anoth­er piece of arti­fi­cial intel­li­gence-gen­er­at­ed music, this one a poly­phon­ic piece trained on Bach chorales and per­formed with the sounds of NSynth.

If you’d like to see how the cre­ation of nev­er-before-heard instru­ments works in a bit more depth, have a look at the demon­stra­tion just above of the NSynth inter­face for Able­ton Live, one of the most DJ-beloved pieces of audio per­for­mance soft­ware around, just above. Hear­ing all this in action brings to mind the moral of a sto­ry Bri­an Eno has often told about the DX7, from which only he and a few oth­er pro­duc­ers got inno­v­a­tive results by actu­al­ly learn­ing how to pro­gram: as much as the prospect of AI-pow­ered music tech­nol­o­gy may astound, the music cre­at­ed with it will only sound as good as the skills and adven­tur­ous­ness of the musi­cians at the con­trols — for now.

Relat­ed Con­tent:

Arti­fi­cial Intel­li­gence Pro­gram Tries to Write a Bea­t­les Song: Lis­ten to “Daddy’s Car”

Arti­fi­cial Intel­li­gence Cre­ativ­i­ty Machine Learns to Play Beethoven in the Style of The Bea­t­les’ “Pen­ny Lane”

Watch Sun­spring, the Sci-Fi Film Writ­ten with Arti­fi­cial Intel­li­gence, Star­ring Thomas Mid­dled­itch (Sil­i­con Val­ley)

Two Arti­fi­cial Intel­li­gence Chat­bots Talk to Each Oth­er & Get Into a Deep Philo­soph­i­cal Con­ver­sa­tion

Based in Seoul, Col­in Mar­shall writes and broad­casts on cities and cul­ture. He’s at work on the book The State­less City: a Walk through 21st-Cen­tu­ry Los Ange­les, the video series The City in Cin­e­ma, the crowd­fund­ed jour­nal­ism project Where Is the City of the Future?, and the Los Ange­les Review of Books’ Korea Blog. Fol­low him on Twit­ter at @colinmarshall or on Face­book.