How Machine Learning Is Generating Strange, New Sounds - ft. Andrew Huang

Listen to Andrew's track "Rainbowgram" on...

Project Magenta is a Google Research project that uses machine learning to create new tools for artists and musicians. One of these tools is NSynth, a neural synthesizer that generates strange, new sounds like “cat flutes,” “beast guitars,” and “screaming 3D printers.” In this video, I explore how NSynth works, and talk with Andrew Huang about his process creating a song entirely out of NSynth generated sounds.

~~~~Magenta Links!~~~~

~~~~Thanks to these amazing people who helped me make this video!~~~~

Also a big thanks to...

:)
Nat & Friends

[natandfriends]

Andrew has now released his full track "Rainbowgram" on a bunch of platforms. You can listen to it here:

[TofuRabbit]

Omg I absolutely love the visuals you put in this video. Even the name tag animations like the one at 2:00 were so charming and clever. Super well done.

[ehwored]

This video combined so many nice things. Nat + Andrew + Ai-Music.

[cesarperez]

Your videos look like they take up a lot of time and they deserve more views.

[IanGrams]

Your nonchalant hair fixing after the strobe dance made me snort 😂 Awesome video! Having prior read about nsynth it was great to learn more about it and hear it in action.

[alxrm]

That strobe dancing sequence is just incredible :0

[paradonym]

Noodle's expression when Nat said "bye": 'you didn't mean that seriously? Of course you don't mean this seriously...'

[snipr]

Came from Andrew Huang aka the guy who makes Music out of anything :)

[gbhall]

Warning strobe light at 9mins if you suffer from photosensitive epilepsy

[justanormalperson]

wow i never learned that much from a video thanks

[NickAc]

"It's basically a flute that meows."

[jakubmarkowski]

Would love to see more videos more often! ;)

[hayksamvelyan]

Man sometimes I remember Lo and I really miss her personality in these videos. I think this channel needs that Ying Ying dynamic that Lo used to add to the mix.

Having said that, these videos are so well put together and are very educational without pushing Google products onto their viewers. I really respect that.

[sshum]

Take ALL the songs in the NewRetroWave channel, jam it in with some unicorn ML and let's see what retro synth track it can produce.

[lunagraysen]

Wished there was warning for the strobe light sequence. Epilepsy is a thing… I personally have it mild but I’m sure this hit other people with it more severe.

[Seegalgalguntijak]

So if you were to combine the bark of a dog and the meow of a cat together as latent vectors with nsynth, what you'd get is the bark of a fox. At least that's how a fox sounds to me. And no, this isn't a "what did the fox say" comment, it was always my observation when I heard one bark that it sounded like the voice of a cat making the sound of a dog's voice. I'd really like to hear nsynth's interpretation of the bark-meow-combination as a comparison to that!

[RagnarRipper]

That is insanely strange and so wonderful.

[RoshnaOmer]

This was amazing! Thanks Nat!
I loved learning about audio manipulation.
Why not talk about NLP and language next time?

[MusicalDistractions]

I don't know why youtube recomended me this video with it's dumb algorithm but I'm really glad it did! One of my favourite videos I've seen in a while.

[jemmapellemma]

Absolutely excellent! This was a joy to experience!
I knew of Andrew Huang but not you. Subbed! ;)