0:00

i'm british undertaker i'm 18

0:03

that i created speak up

0:05

a speech, 84 the paralyzed, whatever

0:07

you plan on doing or whatever you want

0:10

to do you just have to like start i

0:14

do not taking the chance in the first place, then

0:17

i wouldn't be where i am right now

0:21

anyone

0:23

can really do anything as long as you

0:25

have an idea and take

0:27

it off

0:36

wow computer genius generation apart

0:39

at the we talk to young people doing horrible things

0:41

and world of sam i'm your host

0:43

after centimeter cel on the science producer

0:45

a writer and concentrate or with a phd

0:47

and communication i'm here with

0:49

her and saunders shaker who is a senior

0:51

high school student and the inventor as he

0:53

does device that captures the neurological

0:55

signals from brain and translate them

0:57

into words that people are trying to say

1:00

so mockingbird hey okay

1:03

so speak sound like a device from the future

1:05

can give us a brief rundown as to how the

1:07

device works yeah

1:11

the i people can't communicate and

1:13

they have turn to i and see

1:15

tracking devices to be able to speak so

1:17

think of stephen hawking right used

1:20

a device that essentially measured

1:22

is seek twitching that can be extremely

1:24

slow and fatiguing

1:26

way to communicate so with speak

1:28

up doesn't said is a focuses

1:31

on translating your brain

1:33

signals into once you

1:35

want to actually say so even

1:37

though people may be paralyzed their brains

1:40

to sense article signals to the speech system

1:42

and although their muscles may

1:45

be locked in place and can act on those signals

1:47

you can still catch those signals and translate

1:49

them into one person was

1:51

trying say in the first place by using

1:54

some machine learning technology

1:56

the like way in a cab tennis the device

1:58

actually picked up the brain signals

2:00

i'm capturing the electrical signals

2:02

sent he used be system so i'm capturing

2:04

it from your throat oh i'm thinking

2:06

alters on the throw and capturing mg

2:08

signals through this i chose that your brain

2:11

said when your brain is telling your arm

2:13

to move or something it always and he mg signals

2:15

to your biceps and so whenever

2:17

said speak your also moving your muscles

2:20

like your lips and your tongue and

2:22

a specific manner so you're generating

2:24

dmg signals of certain patterns

2:26

to be able to say those words the first place wow

2:29

that sounds hard core how did he

2:31

get into that during

2:33

my sophomore year of high school i was teaching and

2:35

a piece seminar class and

2:37

inner a piece americans i was researching

2:39

how technologies used to

2:42

tackle disability production problem

2:45

and ran into this paper in nineteen eighties

2:47

dad went over some ways

2:49

to create be said

2:51

for people who are powerless and

2:53

one ideas that were highlighted in

2:55

that paper was this idea

2:57

of transiting brain signals into

3:00

the speech for does words

3:02

that someone actually wanted say in the first place

3:04

and so although didn't have the

3:07

ability to do that in the nineteen

3:09

eighties because of computing power

3:11

recently universities like mit

3:13

have been able to predict had species

3:16

the last year when want do want research project

3:18

i was really curious was to how

3:21

this technology worked and wanted create my

3:24

own little sort device of device created created

3:26

costs version and tried to optimize the

3:28

transition algorithm that took these brain

3:30

signals and translated them into words

3:33

an algorithm that you're talking about how

3:35

does that work can give a little bit clarification

3:37

us like higher optimizing this algorithm

3:39

i'm using machine learning technology which

3:41

is essentially where you provide

3:44

a bunch of data had to computer

3:47

and computer is tasked

3:49

with finder understand

3:52

how this data fits together

3:54

and drop patterns within this dataset

3:57

so like for the letter a i would

3:59

have would have of the empty signals and for

4:01

another letter like letter he i would have another

4:04

set of dmg signals and computer

4:06

would go through the status and understand

4:08

that oh is the pattern of the signal

4:10

looks like this than has to be the letter a

4:12

and the pattern of signal looks like this

4:15

than it has to be the letters eve

4:17

is this something that you taught yourself or

4:19

is there a class that you're taking, how did you learn

4:21

to pick up machine learning? and

4:30

the dream was always like, robot rock

4:32

stuff like that but obviously

4:34

never knew how to go about doing that

4:36

kind of stuff and so in seventh

4:39

grade my parents bought me an arduino

4:41

which is like a microcontroller that you

4:43

can program using your laptop to

4:45

like make of l e d blink motor

4:48

turn on and stuff like that and i went through

4:50

some youtube videos at the tammy how to

4:53

use a murder or turn on similarly

4:55

deal it's you make little display and

4:57

stuff like that so already had like the basic

4:59

programming knowledge ended

5:01

up over the summer after sophomore

5:03

year i just picked up a course on coursera

5:06

and went through these machine learning courses

5:08

of said that was interesting and so i learned

5:11

the knowledge necessary to be able to implement

5:13

these machine any synergies

5:15

i wanted kind circle back really click into

5:17

your device when you're trying to teach it's

5:19

different types of vowels to continue

5:21

update the data the for your machine learning

5:24

are you had nothing your own sensors

5:26

year

5:26

own throat while you're doing it so

5:29

those are the painful process i sat

5:31

down my desk i connected does doctors

5:33

to my throat

5:34

and then i said like the letter a

5:36

two hundred times i would save

5:39

those signals on like an sd card then

5:41

i say the letters eve to myself two hundred

5:43

times and took hours sometimes

5:45

signals are no records that i'd like that

5:48

might have as air date again

5:50

and unlike repeat the hopping assess how

5:53

if if like five leaders who

5:56

they want use in times of me

5:58

saying letters myself

6:00

over and over again

6:01

i think sometimes we think of designing

6:03

inventions we glamorize to make

6:05

seem like oh my god look at final product when

6:07

actually is like the mundane this and repetitiveness

6:10

of their back signing can be painful but it's

6:12

the pay out of it actually working

6:14

at the end so while you're doing this process

6:16

whether any setbacks that you have

6:18

was this the most challenging part in developing this

6:20

device

6:21

there's coating aspect because you'd always like

6:23

wanting to errors in ,

6:25

your program something and just just

6:28

out like this random air they do have no

6:30

idea that means the search it off so

6:32

give you bunch of losers on mind none

6:34

of them of out easier to just have to

6:36

like see trying and try to figure out what's

6:38

wrong and then you managed

6:40

you us but then you

6:42

fix that air and then you realize is no other

6:44

air and your code and you go back

6:47

and look at what else

6:49

could be wrong cruises hours of just sitting

6:51

in my computer and know that my dad

6:53

to be sitting the living room watching tv and

6:55

i got them and be like ran into

6:58

the air i don't i do this and i believe

7:00

okay now go back to my room for couple

7:02

hours and then i'd go back to him as good

7:04

as thought that that around to another air

7:06

and i don't have to do this and logistics speaks

7:08

things recent just

7:11

resistance to listen and let me just

7:14

me her that moment that

7:17

an aspect to it those challenging

7:20

was the fact that i didn't know whether

7:22

of my project would working on obviously

7:25

it's been done before so i know it's like

7:28

a possible thing do because researchers

7:30

at mit have been able create such

7:32

device with like high accuracy that

7:34

they translate different words and multitude

7:36

of words and sentences but

7:39

i was using low cost technology

7:41

like sensor that had to

7:43

take my dmg signals causes thirty

7:45

dollars and i bought off amazon so first

7:47

of i don't have that was an accurate enough censor

7:50

to be able be get those

7:52

minute discrepancies minute those signals

7:54

to be able to parse out different

7:57

letters and different also didn't know whether

7:59

or not dogs citing would be

8:01

able to correctly identify whether

8:03

certain signal was flattered a

8:05

with saturday or whatever

8:07

valid was and so i

8:09

was kind of just a blind the hoping would

8:11

work when i made my dataset

8:14

in first place and then ran through my computer

8:16

and through my algorithm is put out accuracy

8:19

value of like sixty percent and that's when knew

8:21

that i was roughly the ballpark of trying get

8:23

to work

8:25

in terms of female hope that even

8:27

though you're using low resources

8:29

you are comparing yourself to

8:31

an institution the has so much more resources

8:34

like mit is that speaks volumes to

8:36

kind of like your resourcefulness and the fact

8:38

they even know your you the quote unquote low quality

8:40

devices from like amazon as her sensors

8:43

that you're still able to produce elise significant

8:46

amount of accuracy athena really cool

8:48

and pretty impressive to think about where

8:51

is your project now you've

8:53

trained it with you five hour was

8:55

you've been able to see that the accuracy

8:57

of thing up for the most part what

8:59

is the next stage what are the

9:01

eat your plans i say the next

9:03

steps to this project is to

9:05

just try and implemented for

9:07

more letters i think the two

9:09

main types are to increases as a

9:11

dataset to be able to classify

9:14

more letters and words and

9:16

sewer add more sensors to be able

9:18

to improve the accuracy what are your

9:20

future plans you're near the end of your high

9:23

school if you have done a

9:25

good amount if

9:25

it can't work with speak up are you

9:27

wanted to continue that out high school i

9:29

said i'm researching more into

9:32

this technology we just cause brain computer

9:34

interfaces

9:36

in college but i do

9:38

plan on continuing this project into

9:40

college where i have like more resources and

9:42

help from other people if you're able to

9:44

kind of continue to progress that's what the impact

9:46

that you're wanting to leave from speak up that's

9:48

a hard question younger students like

9:51

ask me how do do these projects

9:53

or how do you go about thinking

9:55

about this are accomplishing that i

9:57

want to let students know that

10:00

now it's special ability a special talent

10:02

i'm not special kid not a genius

10:06

assist these skills that

10:08

you learn and acquire any one

10:10

can really do anything as long as

10:12

you have the skills to be able

10:14

to execute attack and

10:16

whatever you plan on doing or whatever

10:19

you to do like

10:21

like

10:24

a bad idea for might

10:26

be said for years ago i started programming

10:29

there's no shots that could have done

10:31

by because it does

10:33

for used again i'm again develop the

10:35

skills to smaller projects

10:38

i've been able to acquire knowledge

10:40

necessary to be able

10:42

to have idea and aged

10:45

on are not seek in the sense in

10:47

the first place then first wouldn't

10:49

be where i am right now it's fast yeah

10:52

i think the seeking of the skills that you pick

10:54

up the

10:54

company's of hone in on your device

10:56

smithy feel like what i get from you is

10:58

your resourcefulness we have we have those resources

11:01

online for the most part free available

11:03

the taken advantage of the things we have access

11:05

to

11:06

the kind of like continue to for that curiosity

11:08

and spark that desire to once continued

11:11

to learn give you the

11:13

internet's all disadvantage and so

11:15

i think that's really awesome i use my and

11:17

it doesn't mean have to be genius think the things

11:20

is all about having the curiosity

11:22

and the resources to continue the

11:25

very insightful conversation

11:28

about you're working as the interface of

11:30

kind of like this new wave of technology

11:32

and using it's you continue to in

11:34

order for a to be used for paralyzed

11:36

folks think it's awesome again

11:39

to bike your genius but

11:41

guess like your dedication the wanting is

11:44

free and kind of cause an impact

11:46

in white girl designing with your programming

11:49

and i hope you continue to developer skill

11:51

than i hope to see what's next for you in teachers

11:53

the thank you so much room for joining us today

11:55

i thank for having me thank

11:57

you so much for listening this week's episode the

12:00

reason nectar to kinda nice here

12:02

and next are incredible sunsets

12:13

funding for the genius generation comes

12:15

from the arthur vining davis foundations

12:18

investing in our common future support

12:21

for tracks comes from the corporation well the broadcasting

12:26

this is tracks from p

12:28

r x