[AI Experiments: The Infinite Drum Machine]
Hey, I am kyle.
我是Annie 嗨 我是Yotam
I am Annie.Hi, I am Yotam.
[coders / musicians]
This is something we made called infinite drum machine.
It’s basically an experiment to see how machine-learning might organize everyday sounds.
So anyone who’s ever had to work with a lot of sound can see why this is hard.
For example,we can all hear that this is similar to this
but totally different from this.
But it’s hard to articulate exactly why, because sounds are complicated.
But this is kind of thing that machine learning is really good at.
So we ran a little test.
We played thousands of sounds for the computer.
We didn’t give it descriptions or tags
or any hints on how people usually organized sound.
We wanted instead to see if could listen to them
and automatically learn its own way to organize them.
This experience uses a technique called t-SNE.
First the computer creates a fingerprint from each sample,
which is like an image or a set of numbers that represents that particular sound.
Then t-SNE compares all the fingerprints and places similar sounds close together.
One way to think about t-SNE is that
it’s taking these fingerprints from high dimensional space,
which means the space is more than three dimensions
and reducing them in this case to two dimensions
so that we can visualize it.
And our test basically worked.
This is a t-SNE that we got.
We put it on the web so anyone can play with it.
You get these neighborhoods of similar sounds
that you can explore by dragging around.
And may beat with them by using this drum machine.
We are hoping people find it fun,
whether you are more into the machine learning side
or more into making music.
You can play with it and get the code at G.CO/AIEXPERIMENTS.