Rupal Patel: Synthetic voices, as unique as fingerprints

Translator: Nika Kotnik Reviewer: Metka Hercog

I'd like to talk today

Danes bi rada govorila

about a powerful and fundamental aspect

o močnem in bistvenem vidiku

of who we are: our voice.

tega kar smo. O našem glasu.

Each one of us has a unique voiceprint

Vsak od nas ima unikaten odtis glasu

that reflects our age, our size,

ki odseva našo starost, našo velikost,

even our lifestyle and personality.

celo naš stil življenja in osebnost.

In the words of the poet Longfellow,

V besedah pesnika Longfellowa:

"the human voice is the organ of the soul."

"Človeški glas je glasbilo duše."

As a speech scientist, I'm fascinated

Kot znanstvenico govora me navdušuje

by how the voice is produced,

nastanek zvoka

and I have an idea for how it can be engineered.

in imam idejo, kako bi ga lahko naredili.

That's what I'd like to share with you.

To bi rada delila z vami.

I'm going to start by playing you a sample

Začela bom s predvajanjem vzorca glasu,

of a voice that you may recognize.

ki ga boste morda prepoznali.

(Recording) Stephen Hawking: "I would have thought

(Posnetek) Stephen Hawking: Menil bi,

it was fairly obvious what I meant."

da je bilo popolnoma očitno, kaj sem mislil."

Rupal Patel: That was the voice

Rupal Patel: To je bil glas

of Professor Stephen Hawking.

profesorja Stephena Hawkinga.

What you may not know is that same voice

Ne veste pa, da bi isti glas lahko

may also be used by this little girl

uporabljala ta deklica,

who is unable to speak

ki ne more govoriti

because of a neurological condition.

zaradi nevrološke okvare.

In fact, all of these individuals

Pravzaprav, vsi ti posamezniki

may be using the same voice,

bi lahko uporabljali ta glas,

and that's because there's only a few options available.

ker je na voljo zelo malo možnosti.

In the U.S. alone, there are 2.5 million Americans

Samo v ZDA, je okrog 2,5 milijona Američanov

who are unable to speak,

ki ne morejo govoriti,

and many of whom use computerized devices

in mnogi izmed njih uporabljajo računalniške naprave

to communicate.

za komunikacijo.

Now that's millions of people worldwide

To so milijoni ljudi po vsem svetu,

who are using generic voices,

ki uporabljajo generični glas

including Professor Hawking,

kot profesor Hawking,

who uses an American-accented voice.

ki uporablja glas z ameriškim naglasom.

This lack of individuation of the synthetic voice

To pomanjkanje individualnosti sintetičnega glasu

really hit home

me je res zadelo,

when I was at an assistive technology conference

ko sem bila na konferenci tehnologije za pomoč

a few years ago,

pred nekaj leti,

and I recall walking into an exhibit hall

in spominjam se, da sem stopila v razstavno dvorano

and seeing a little girl and a grown man

in videla, kako se deklica in odrasel moški

having a conversation using their devices,

pogovarjata s svojima napravama,

different devices, but the same voice.

različne naprave, a enak glas.

And I looked around and I saw this happening

Pogledala sem okrog in to se je dogajalo

all around me, literally hundreds of individuals

vse okrog mene, na stotine posameznikov

using a handful of voices,

je uporabljalo samo nekaj glasov,

voices that didn't fit their bodies

glasov, ki se niso skladali z njihovimi telesi

or their personalities.

ali osebnostmi.

We wouldn't dream of fitting a little girl

Niti v sanjah ne bi deklici namestili

with the prosthetic limb of a grown man.

proteze uda za odraslega moškega.

So why then the same prosthetic voice?

Zakaj torej enak protetični glas?

It really struck me,

Res me je zadelo,

and I wanted to do something about this.

in nekaj sem želela storiti glede tega.

I'm going to play you now a sample

Sedaj vam bom predvajala vzorec

of someone who has, two people actually,

nekoga, ki ima, dva človeka pravzaprav,

who have severe speech disorders.

ki imata hude motnje govora.

I want you to take a listen to how they sound.

Želim, da slišite, kako zvenita.

They're saying the same utterance.

Izgovarjata isti zvok.

(First voice)

(Prvi glas)

(Second voice)

(Drugi glas)

You probably didn't understand what they said,

Najbrž niste razumeli, kaj sta rekla,

but I hope that you heard

ampak upam, da ste slišali

their unique vocal identities.

njuni unikatni glasovni identiteti.

So what I wanted to do next is,

Potem sem hotela

I wanted to find out how we could harness

izvedeti, kako bi lahko izkoristili

these residual vocal abilities

te preostale vokalne zmožnosti

and build a technology

in zgradili tehnologijo,

that could be customized for them,

ki bi jim bila prilagojena,

voices that could be customized for them.

glasovi, ki bi jima bili prilagojeni.

So I reached out to my collaborator, Tim Bunnell.

Povezala sem se s svojim sodelavcem, Timom Bunnelom.

Dr. Bunnell is an expert in speech synthesis,

Dr. Tim Bunnell je strokovnjak za sintezo govora,

and what he'd been doing is building

in on ustvarja

personalized voices for people

personalizirane glasove za ljudi

by putting together

s sestavljanjem

pre-recorded samples of their voice

prej posnetih vzorcev njihovega glasu

and reconstructing a voice for them.

in rekonstrukcijo glasu zanje.

These are people who had lost their voice

To so ljudje, ki so izgubili svoj glas

later in life.

kasneje v življenju.

We didn't have the luxury

Mi nismo imeli te sreče,

of pre-recorded samples of speech

da bi imeli prej posnete vzorce govora

for those born with speech disorder.

ljudi, ki so rojeni z motnjo govora.

But I thought, there had to be a way

Ampak pomislila sem, da gotovo obstaja način,

to reverse engineer a voice

da z obratnim inženiringom dobimo glas

from whatever little is left over.

iz tistega, kar je še ostalo.

So we decided to do exactly that.

Točno to smo se odločili storiti.

We set out with a little bit of funding from the National Science Foundation,

Začeli smo z malo financiranja Nacionalne znanstvene fundacije,

to create custom-crafted voices that captured

da bi ustvarili po meri narejene glasove, ki bi ujeli

their unique vocal identities.

njihove edinstvene glasovne lastnosti.

We call this project VocaliD, or vocal I.D.,

Temu projektu pravimo VocaliD, ali vocal I.D.,

for vocal identity.

vokalna identiteta.

Now before I get into the details of how

Preden grem v podrobnosti

the voice is made and let you listen to it,

izdelave glasu in vam ga pustim poslušati,

I need to give you a real quick speech science lesson. Okay?

vam moram res hitro razložiti osnove znanosti glasu, v redu?

So first, we know that the voice is changing

Torej, vemo, da se glas dramatično

dramatically over the course of development.

spreminja med razvojem.

Children sound different from teens

Otroci zvenijo drugače od najstnikov,

who sound different from adults.

ki zvenijo drugače kot odrasli.

We've all experienced this.

Vsi smo to izkusili.

Fact number two is that speech

Drugo dejstvo je, da je govor

is a combination of the source,

kombinacija vira,

which is the vibrations generated by your voice box,

to so vibracije, ki jih proizvaja vaš glasovni aparat,

which are then pushed through

ki jih potem potisnemo

the rest of the vocal tract.

skozi preostali vokalni trakt.

These are the chambers of your head and neck

To so votline v vaši glavi in vratu,

that vibrate,

ki vibrirajo,

and they actually filter that source sound

in dejansko filtrirajo ta izvorni zvok,

to produce consonants and vowels.

da ustvarijo soglasnike in samoglasnike.

So the combination of source and filter

S kombinacija izvora in filtra

is how we produce speech.

mi proizvajamo zvok.

And that happens in one individual.

In to se zgodi v enem posamezniku.

Now I told you earlier that I'd spent

Prej sem vam povedala, da sem preživela

a good part of my career

dober del svoje kariere,

understanding and studying

da sem preučevala in skušala razumeti

the source characteristics of people

izvorne lastnosti ljudi

with severe speech disorder,

s hudimi motnjami govora,

and what I've found

in odkrila sem,

is that even though their filters were impaired,

da čeprav so bili njihovi filtri poškodovani,

they were able to modulate their source:

so lahko uravnavali svoj vir:

the pitch, the loudness, the tempo of their voice.

višino, glasnost, tempo svojega glasu.

These are called prosody, and I've been documenting for years

To se imenuje prozodija, in že leta dokumentiram,

that the prosodic abilities of these individuals

da so prozodične sposobnosti teh posameznikov

are preserved.

ohranjene.

So when I realized that those same cues

Ko sem spoznala, da so te iste iztočnice

are also important for speaker identity,

prav tako pomembne za identiteto govorca,

I had this idea.

sem dobila idejo.

Why don't we take the source

Zakaj ne vzamemo vira

from the person we want the voice to sound like,

nekoga, čigar glas si želimo,

because it's preserved,

ker je ohranjen,

and borrow the filter

in si filter sposodimo

from someone about the same age and size,

od nekoga istih let in velikosti,

because they can articulate speech,

ker lahko artikulirajo svoj govor,

and then mix them?

in jih potem pomešamo?

Because when we mix them,

Ker ko jih pomešamo,

we can get a voice that's as clear

dobimo glas, ki je tako jasen

as our surrogate talker --

kot naš govornik darovalec -

that's the person we borrowed the filter from—

to je oseba, pri kateri smo si izposodili filter -

and is similar in identity to our target talker.

in je v identiteti podoben našemu govorcu.

It's that simple.

Tako preprosto je.

That's the science behind what we're doing.

To je znanost za našim delom.

So once you have that in mind,

Ko imaš enkrat to v mislih,

how do you go about building this voice?

kako potem zgradiš ta glas?

Well, you have to find someone

No, najti moraš nekoga,

who is willing to be a surrogate.

ki je pripravljen biti darovalec.

It's not such an ominous thing.

To ni tako zlovešča stvar.

Being a surrogate donor

Biti darovalec

only requires you to say a few hundred

zahteva samo izgovorjavo nekaj sto

to a few thousand utterances.

do nekaj tisoč glasov.

The process goes something like this.

To proces poteka nekako takole.

(Video) Voice: Things happen in pairs.

(Video) Glas: Stvari se dogajajo v parih.

I love to sleep.

Rad spim.

The sky is blue without clouds.

Nebo je modro brez oblakov.

RP: Now she's going to go on like this

RP: Takole bo šla naprej

for about three to four hours,

približno tri do štiri ure,

and the idea is not for her to say everything

in ideja ni, da izreče vse,

that the target is going to want to say,

kar bo hotela izreči prejemnica,

but the idea is to cover all the different combinations

ampak je ideja, da pokrijemo vse različne kombinacije

of the sounds that occur in the language.

zvokov, ki se lahko pojavijo v jeziku.

The more speech you have,

Več govora kot imaš,

the better sounding voice you're going to have.

lepše zveneč glas boš imel.

Once you have those recordings,

Ko imamo te posnetke,

what we need to do

jih moramo

is we have to parse these recordings

razčleniti

into little snippets of speech,

v majhne drobce govora,

one- or two-sound combinations,

kombinacije enega ali dveh zvokov,

sometimes even whole words

včasih celo cele besede

that start populating a dataset or a database.

da začnemo zapolnjevati set podatkov ali podatkovno bazo.

We're going to call this database a voice bank.

Tej bazi podatkov bomo rekli glasovna banka.

Now the power of the voice bank

Moč glasovne banke

is that from this voice bank,

je v tem, da iz nje

we can now say any new utterance,

lahko izgovorimo karkoli,

like, "I love chocolate" --

recimo: "Rada imam čokolado" -

everyone needs to be able to say that—

vsakdo mora biti sposoben izreči to -

fish through that database

pobrskamo po bazi podatkov

and find all the segments necessary

in najdemo vse potrebne segmente

to say that utterance.

za izgovorjavo tega.

(Video) Voice: I love chocolate.

(Video) Glas: Rada imam čokolado.

RP: So that's speech synthesis.

RP: To je sinteza govora.

It's called concatenative synthesis, and that's what we're using.

To je konkatenativna sinteza, in to mi uporabljamo.

That's not the novel part.

To ni nekaj novega.

What's novel is how we make it sound

Nov del je, kako naredimo, da zveni

like this young woman.

kot ta mlada ženska.

This is Samantha.

To je Samantha.

I met her when she was nine,

Spoznala sem jo, ko je imela 9 let,

and since then, my team and I

in od takrat ji z ekipo

have been trying to build her a personalized voice.

poskušamo zgraditi njen osebni glas.

We first had to find a surrogate donor,

Najprej smo morali najti nadomestnega donorja,

and then we had to have Samantha

potem je morala Samantha izgovoriti

produce some utterances.

nekaj zlogov.

What she can produce are mostly vowel-like sounds,

Lahko proizvaja predvsem samoglasnike,

but that's enough for us to extract

a to je bilo za nas dovolj, da smo lahko pridobili

her source characteristics.

njene značilnosti.

What happens next is best described

Kar se zgodi potem, najlepše opiše

by my daughter's analogy. She's six.

analogija moje hčere. Šest let ima.

She calls it mixing colors to paint voices.

Temu pravi, da mešamo barve, za risanje glasov.

It's beautiful. It's exactly that.

Prelepo je. Točno tako je.

Samantha's voice is like a concentrated sample

Samanthin glas je kot koncentriran vzorec

of red food dye which we can infuse

rdečega barvila za hrano, ki ga lahko damo

into the recordings of her surrogate

v posnetke njenega darovalca,

to get a pink voice just like this.

da dobimo tak rožnat glas.

(Video) Samantha: Aaaaaah.

(video) Samantha: Aaaaah.

RP: So now, Samantha can say this.

RP: Sedaj, lahko Samantha reče tole.

(Video) Samantha: This voice is only for me.

(Video) Samantha: Ta glas je samo zame.

I can't wait to use my new voice with my friends.

Komaj čakam, da uporabim svoj glas s svojimi prijatelji.

RP: Thank you. (Applause)

RP: Hvala. (Aplavz)

I'll never forget the gentle smile

Nikoli ne bom pozabila nežnega nasmeška,

that spread across her face

ki se je razširil čez njen obraz,

when she heard that voice for the first time.

ko je prvič slišala ta glas.

Now there's millions of people

Tu so milijoni ljudi

around the world like Samantha, millions,

po celem svetu, ki so kot Samantha, milijoni,

and we've only begun to scratch the surface.

in šele začeli smo.

What we've done so far is we have

Sedaj imamo

a few surrogate talkers from around the U.S.

nekaj nadomestnih govorcev po celih Združenih državah Amerike,

who have donated their voices,

ki so darovali svoj glas,

and we have been using those

in mi smo jih uporabljali,

to build our first few personalized voices.

da smo zgradili naše prve personalizirane glasove.

But there's so much more work to be done.

A veliko dela je še pred nami.

For Samantha, her surrogate

Za Samantho, njen donor

came from somewhere in the Midwest, a stranger

prihaja nekje iz srednjega zahoda, tujec,

who gave her the gift of voice.

ki ji je dal dar glasu.

And as a scientist, I'm so excited

Kot znanstvenica, sem tako vznemirjena,

to take this work out of the laboratory

da smo vzeli to delo iz laboratorija

and finally into the real world

v realni svet,

so it can have real-world impact.

kjer ima lahko realen vpliv.

What I want to share with you next

Z vami bi rada delila,

is how I envision taking this work

kako bi popeljala to delo

to that next level.

na naslednjo raven.

I imagine a whole world of surrogate donors

Predstavljam si cel svet nadomestnih darovalcev,

from all walks of life, different sizes, different ages,

vseh vrst, različnih velikosti, različne starosti,

coming together in this voice drive

kako se srečujejo v tej bazi glasov,

to give people voices

da bi dali ljudem glasove,

that are as colorful as their personalities.

ki so tako barviti kot njihove osebnosti.

To do that as a first step,

Prvi korak je bila

we've put together this website, VocaliD.org,

postavitev spletne strani, VocaliD.org,

as a way to bring together those

da bi zbrali tiste,

who want to join us as voice donors,

ki bi se nam radi pridružili kot darovalci glasov,

as expertise donors,

kot izkušeni darovalci,

in whatever way to make this vision a reality.

na kakršenkoli način, da bi ta načrt postal resničnost.

They say that giving blood can save lives.

Pravijo, da darovanje krvi rešuje življenja.

Well, giving your voice can change lives.

No, če posodite svoj glas, lahko spremenite življenja.

All we need is a few hours of speech

Vse, kar potrebujemo, je nekaj ur govora

from our surrogate talker,

našega nadomestnega govorca

and as little as a vowel from our target talker,

in samo samostalnik našega govorca,

to create a unique vocal identity.

da ustvarimo unikatno glasovno identiteto.

So that's the science behind what we're doing.

To je znanost za našim delom.

I want to end by circling back to the human side

Zaključila bi s človeško platjo,

that is really the inspiration for this work.

ki je resnični navdih za to delo.

About five years ago, we built our very first voice

Pred petimi leti, smo zgradili naš prvi glas

for a little boy named William.

za majhnega fanta po imenu William.

When his mom first heard this voice,

Ko je njegova mama prvič slišala ta glas,

she said, "This is what William

je rekla: "Tako bi William zvenel,

would have sounded like

had he been able to speak."

če bi lahko govoril."

And then I saw William typing a message

Videla sem Williama, kako tipka sporočilo

on his device.

na svoji napravi.

I wondered, what was he thinking?

Spraševala sem se, le o čem razmišlja?

Imagine carrying around someone else's voice

Predstavljajte si, da nosite glas nekoga drugega

for nine years

devet let

and finally finding your own voice.

in končno najdete svoj glas.

Imagine that.

Predstavljajte si.

This is what William said:

To je rekel William:

"Never heard me before."

"Sebe pa še nisem slišal."

Thank you.

Hvala.

(Applause)

(Aplavz)