of people whose speech is hard for others to understand.

They're not used in training

the speech recognition model.

I mean, the game is to record things

and then have it recognize things that you say that aren't

in the training set.

Dimitri recorded 15,000 phrases.

It wasn't obvious that this was going to work.

He just sat there and he kept recording.

[Dimitri speaking]

You can see that it's possible

to make a speech recognizer work for Dimitri.

It should be possible to make it work for many people.

Even people who can't speak

because they've lost the ability to speak.

The work that Shanqing has done on,

you know, voice utterances.

[hum]

From sounds alone you can communicate.

But there might be other ways of communicating.

Most people with ALS end up using

an on-screen keyboard and having to type each

individual letter with their eyes.

For me, communicating is sloooooooow.

Steve might crack a joke

and it's related to something that happened,

you know, a few minutes ago.

The idea is to create a tool so that

Steve can train machine learning models himself

to understand his facial expressions.

[basketball crowd cheering]

[air horn sounds]

Michael: [laughs] it works!

to be able to laugh to be able to cheer to be to boo,

things that seem maybe superfluous,

but actually are so core to being human.

I still think this is only the tip of the iceberg.

We're not even scratching the surface yet of what is possible.

If we can get speech recognizers to work

with small numbers of people,

we'll learn lessons which we can then combine to build something

that really works for everyone.

♪♪