Latest / Identity at the Center / #192 - Voice Biometrics with Mel Dingman

Transcript

#192 - Voice Biometrics with Mel Dingman · Identity at the Center

0:09
You're listening to the identity of the sender podcast.
0:12
This is a show that talks about identity and access management
0:15
and making sure you know who has access to what let's get
0:18
started. Welcome to the identity the
0:26
center podcast I'm Jeff and that's Jim hey, Jim khf.
0:29
How are, you know, not so bad yourself.
0:32
Good good. I've got a lot on my mind these
0:34
days, there's a lot going on in the world right now and I think
0:39
a couple of the key topics for Our Generation to kind of figure
0:45
out and I don't know if we'll figure it out, but it's one of
0:47
the biggest undertones is around free speech and privacy.
0:53
And the reason why Why I tied those two together is because I
0:58
think that there's some kind of trade-off, right.
1:00
There's is not most people aren't black and white about the
1:05
topic. Like I want 100% free speech, no
1:08
controls at all. Because the other side of that
1:11
is there some speech that can cause you know, Clear and
1:16
Present dangers and the same thing with privacy it's like yes
1:21
I'm willing to give up some privacy to get some services.
1:25
But I'm not willing to go up on my privacy and it's not like,
1:29
I'm not willing to give up any data about myself.
1:32
And so, you know what I wanted to pose to you is where are you
1:38
on the Spectrum, in terms of privacy, how much of your
1:41
private data? Are you willing to give up to
1:44
get Services? It depends on the value of the
1:47
service. And then, you know, Google Maps
1:50
for example. I'm okay with location on that
1:53
because it's so helpful for Driving when it comes to
1:56
navigation, sharing location with friends or family or things
1:59
like that, I don't know if it's necessarily a black and white
2:03
decision because every service has its own value, they might
2:08
think it's more important, that it really is to me as a user.
2:12
And there really isn't a lot of discussions sometimes on what
2:14
you can. And can't, you know, switch on
2:16
or off from a privacy standpoint.
2:19
It generally the services are like, well it's either all or
2:21
nothing or more than you'd want to give up.
2:24
Exchange for that service. So I would say I'm pretty
2:26
average sure I give up some modicum of privacy but I try not
2:32
to be too crazy about it. Just you know hit allow on
2:34
everything that pops up on my phone you know asking for things
2:36
like oh does it does my calculator really need to take
2:40
pictures? Probably not.
2:42
So yeah that's that's something here.
2:44
More on like the Android side of things Android apps doing like
2:46
accept the permissions and and things like that but I think
2:51
that's kind of where I fall on it.
2:52
At least from my perspective. Yeah, I kind of feel like that
2:57
the next question is, like, what is the role of government in
3:01
terms of regulations for how much privacy should be
3:08
controlled by those laws and regulations?
3:11
I think, you know, in your PPI GDP are much more restrictive in
3:15
terms of privacy and they wind up, you know, pulling big Tech
3:19
into court and trying to control it a lot more and the United
3:23
States. It's less Rolled and kind of
3:26
Taken like a state-by-state approach, which makes things
3:30
very complex for companies to figure out.
3:33
Okay, you know, how do I deal with these things?
3:36
But I think one of the things to me, that's most annoying when it
3:40
comes to privacy, is things like when you go to a website and it
3:44
says, except our privacy policy. In other words, you cannot get
3:48
access to their service, their online service, unless you
3:54
accept. Kept the privacy policy and then
3:56
you go and you click into it and it's like, do I really want to
3:59
read this? And for some Services, you
4:03
really kind of don't have an option.
4:05
So if you have to go online to pay your utility bill or to
4:09
manage your utility bill, you know, you pretty much just have
4:12
to accept that privacy policy. I kind of the option you can
4:18
show up that at City Hall with a Jar, full of pennies and path,
4:22
that way, right? I mean you're you're Oh, making
4:25
a choice though, right? If you want to go online and
4:26
have like, online bill pay. That is a choice, you're making.
4:30
But as I don't have to share, is that, is that fair or something?
4:37
But I think, no, I think that's the root of the discussion.
4:40
I think, what do we want? This is our society
4:44
collectively, right? What do we want from, you know,
4:50
these Services, right? I mean, you know, just because
4:54
ConEd might be my electric company.
4:56
They didn't invent electricity and they don't.
5:00
And I have, I have to have electricity.
5:04
That's a certain scenario and I picking up the utility scenario
5:08
because you really don't have much choice if you want to have
5:12
electric in your home, but to use their services.
5:15
And yes, you could go down there with rolls of pennies to pay for
5:19
the service, but that's not realistic.
5:21
So in other words, your Kind of forced by.
5:25
You want to have a normal life to accept privacy policy and
5:28
then they can they can take your data now, we're talking about
5:32
services like Facebook or something like that.
5:35
I think now you've entered a different territory because
5:39
that's truly an optional service.
5:42
In most cases, there are some cases where maybe it's not, but
5:47
you know, this is Jim McDonald's own idea, right?
5:50
And I'm going to throw it out there that I think there should
5:53
be Classification standards for privacy policies and a company
5:59
should say should pick one of those privacy policy.
6:03
So, rather than saying you need to read our privacy policy and
6:06
they could stick one sentence somewhere in the privacy policy
6:10
that you don't like that. They have to comply to one of
6:15
whatever it is, 45 standard, privacy policies and then most
6:20
people can kind of understand. Okay, what are those four or
6:22
five privacy policies? And they come to can say, we're
6:25
following number 4. I mean, I think a standardized
6:27
privacy policy would be helpful, but I don't think it's conducive
6:31
to the way that the u.s. is run. You'd have 50% of the population
6:35
but yes, 50% pappu say, no and nothing would happen.
6:39
So I do think I do think maybe it is somewhat crazy but not
6:43
because it's like a crazy idea just because that's the state of
6:46
government in the u.s. at this point that's the way I'm looking
6:49
at it. Yeah.
6:50
Well that is pretty much the state of everything right now,
6:53
right? But imagine if you had to like a
6:55
standardized template, right? Sort of like a Bill of Rights
6:58
for a consumer to be able to easily understand what it is
7:03
that they're getting in exchange for what it is.
7:05
They're giving away for that right?
7:07
Instead of you have like these stupid and user license
7:10
agreement that pop up that are written and you know, legalese
7:14
and lawyer speak and hide all kinds of weird terms and
7:17
conditions on the inside that, oh, you can opt out, but it's on
7:20
page. 74 this tiny little checkbox right or you Have to
7:24
like, write in a letter. It's like, opt out at something,
7:27
like, I'd be totally in favor of like, simplifying that quite a
7:31
bit, but this is a very litigious Society.
7:34
Remember, like all the years, you've been an it and how many
7:38
eula's end-user license agreements, you've clicked
7:42
accept and didn't read, you've agreed to those contracts and I
7:47
think with privacy policies, the same thing, happens all the
7:50
time. We just don't have time to read
7:53
everyone. Well you know what losing you
7:57
know, we've been talking about MFA, right?
7:58
And my face, Pam, and this fatigue that comes with, you
8:02
know, always having to like approve these messages on my
8:04
phone to push notifications. Like, I know it's not me and it
8:07
was not me trying to get in but invariably some percentage of
8:10
population Just Hits allowed because they get tired of it.
8:13
That is exactly what we do with you was like 20 years ago at
8:17
some point, we got tired of it and we just clicked accept and
8:20
we want it right on through. So now we're doing the same
8:22
thing with I know what, LMFAO fatigue or push notification
8:25
fatigue, or whatever it is, but we've talked about it a few
8:27
times on the show. Why don't we pivot?
8:30
Well, let's I want to get to our main topic, which is voice
8:33
Biometrics a couple of housekeeping things.
8:35
Before we do get started, though this is a second last show of
8:38
the year for us. So we're going to take a few
8:40
weeks off for winter. Break 2,700 minutes of podcasts
8:45
went out this year, so I think we've earned a little bit of
8:47
time off to kind of regroup. We are working on some pretty
8:50
cool stuff for 2023. So we'll hit episode Red pretty
8:54
early on I think probably like February maybe March at the
8:57
latest I would imagine at this point we got some conference
9:00
things that were working on so definitely, you know, people
9:04
stay tuned with that hit that subscribe or follow button.
9:07
That totally helps us out, you know, it's a free way to
9:09
basically support the show so that is highly encouraged.
9:14
I've got to tell you a quick story.
9:15
So I was in the airport the other night and I was talking
9:18
with Wayne and somebody was like like He's dropping on her
9:24
conversation. I mentioned to Wayne about the
9:27
the podcast and so this guy turns around sarcastically off
9:32
the podcast. I spent like 15 minutes
9:35
explaining the podcast and he was like all he wanted to find
9:39
it on his podcast app and everything.
9:43
He's like, I'm gonna listen, I'm gonna listen, I'm like all right
9:45
great leave us a five-star rated.
9:46
He's like, well I'll have to listen to podcast first.
9:49
Well, yeah, I was like, oh yeah. Well, you just took him for like
9:53
15 minutes. I time, dude.
9:54
I think you owe me something. Yeah, leave a rating.
9:58
Five stars is always preferred. You know, if you hate it you say
10:00
I really hate it a lot. It was five stars worth.
10:02
I hate that's fine too, right? I mean, either way is good, but
10:06
yeah, yeah, follow subscribe, so pick.
10:08
That's kind of issue. We have two stats last time,
10:10
but, like, less than half of people who routinely turned in
10:12
or actually subscribe to just tune in and hear our drivel
10:16
every week. So why don't we get to voice
10:18
Biometrics? Because they think that's kind
10:20
of where we want to take the take today's conversation.
10:24
Especially some of the areas around privacy concerns.
10:27
That kind of Jim brought up, but voice is a big part of that as
10:29
well, but I think just the technology in general is pretty
10:31
cool. I actually rolled it out two
10:34
decades now. Well, 15 years ago, let's call
10:36
it. It's kind of my first foray into
10:39
that area and I'm really happy and excited that we're able to
10:41
have our guests on with us. She is melting men.
10:44
She's a program director with Booz Allen Hamilton and a voice
10:47
biometric and an identity Zealot.
10:49
So welcome to the show, Mel, thanks for having me.
10:53
Yeah, so, You for taking the time, it's a Friday afternoon
10:56
and this is going to go live. Mondays is going to be kind of a
10:58
quick turnaround for us. And I think one of the things
11:00
that we always like to find out the first time we have someone
11:02
to show is sort of their career Journey especially when it comes
11:05
to sort of the identity work and maybe even specifically The
11:08
Voice Biometrics tough, which you've really kind of
11:10
specialized in it sounds like how did you get to where you
11:13
are? Is this, is this identity and
11:16
Biometrics area? Something that you chose?
11:18
Or did it Choose You? Little bit of both.
11:22
You know, I was in the identity space.
11:24
I'm a lot older, I think than you guys but I started my career
11:28
about 33 years ago and almost immediately jumped in.
11:33
After my webcasting phase don't judge.
11:37
It was a good one back then but then I kind of got my foray into
11:41
IDM identity, access management and I loved it but the more I
11:46
got into it the more I was able to see through Less than always
11:52
legal means the user names and passwords and eat including MFA.
11:58
They just start cutting. They're not they're so easy to
12:02
break and I don't mean that for everybody but the hacker
12:06
community in the black hats around identity it's super easy
12:10
to break pretty much anything that's out there today until you
12:15
talk about Biometrics and you know obviously iPhone or Apple.
12:20
Came out pretty staunchly with facial Rec.
12:25
I know I'm sure every one of us is done the facial dance or
12:28
you're trying to get the face and then you have to do the iris
12:31
dance and you know, it just wasn't, it wasn't feasible and
12:35
mass use and so VB voice Biometrics became hot for me.
12:43
As I saw that there was a lot more that could be done with
12:47
voice then just Identification. So obviously that's the impetus
12:54
behind the explosion of VB, but where science has at the science
13:01
Community has evolved it got super sexy.
13:04
So I rode that I'm all about the super sexy.
13:07
So yeah, we are absolutely and You know, I know we're going to
13:12
get super deep maybe we can start a look, you know, at the
13:16
basics of kind of, for those who aren't familiar with how voice
13:21
Biometrics Works, how does it work?
13:23
But then take us all the way to that privacy topic, that was
13:27
bringing up, like, is my Amazon Alexa spying on me, but let's
13:31
not go there yet. Let's start with the, you know,
13:33
has Voice. Bio generally work.
13:36
So spoiler alert. Yes, it is.
13:39
Secondly VB is it's a very slick process.
13:46
There's lots of variations and how you can deploy VB.
13:50
There are lots of various models, there's a lot of
13:54
different approaches, there's just all kinds of options
14:00
depending on a multitude of factors.
14:01
As you would imagine with any technology you got to understand
14:05
the business requirements that are supporting the operational
14:07
or the mission goals that VB needs to support.
14:11
Port. And then based on that, there
14:12
are actually options now although you wouldn't think so,
14:15
given a lot of the banks and such or still retaining audio
14:19
and they're still leveraging capabilities that are almost as
14:24
old as I am. And so things have evolved.
14:29
If anyone in the fence serve Community is paying attention.
14:33
Things have evolved. Take a second.
14:35
Look at where VB and and Biometrics are today for
14:39
identity management and fraud. Prevention.
14:42
Okay, so you said the Alexa is spying on me.
14:46
What does that mean? Are they just trying to gather?
14:52
Okay, this person needs to go and buy some new, you know, some
14:57
new cups and coffee mugs or they actually like, looking for
15:03
things and Reporting them to the police.
15:07
Yeah, so I mean they're obviously capabilities in the
15:11
intelligence community and the federal government levels that
15:15
are looking for Words keywords that, you know, your phraseology
15:20
and the mix context in which it's uttered.
15:26
But that's not really the job of Alexa as you guys were talking
15:32
about earlier, the whole privacy bit and there's a lot of trust
15:35
right now. On the containers or the the
15:41
holders of your private, your pii, your personal, identifying
15:47
information and your you got to trust and right?
15:49
And it's even devoid of having some sort of, Bill of Rights for
15:53
consumers, or what have you you literally could end up as the
15:57
South Park episode, where their Human Centipede or you are truly
16:02
being protected by the audio and the use of it that's being
16:05
captured. Mel, I kind of Wonder like, you
16:10
know, I think one of the voice applications I use the most is
16:14
Siri right? The Apple version of boy's voice
16:19
recognition and I can't believe how bad it is sometimes.
16:23
So it is what does that mean indication of?
16:25
I mean is an indication of like it could be improved, or is it
16:29
an indication of just like some of the basics of different
16:33
people talking differently? So it's the it's their Universal
16:38
Baseline model which is one of the it's like the transmission
16:43
of voice Biometrics, right? And so the more understands and
16:47
how they are, if they are, which I'm sure they are.
16:51
But I don't have specifics. But as long as they're applying
16:53
Ai and ml outside of what a white paper refers to it, as if
16:58
they're genuinely applying learning capabilities to their
17:02
Baseline construct, then that will get Mm better over time but
17:09
it's further than where they are, right?
17:10
Because, yeah, it's pretty bad. You shouldn't have to repeat
17:13
yourself. Their capabilities that are
17:16
being used in the Department of Defense today and in the
17:20
Department of Homeland Security, where lives depend on the
17:25
efficacy of that voice capture in the translation of it.
17:29
So the technology is there, it just has to be committed to by
17:35
the Vidor of the service and I don't want to, I'm not slamming
17:39
them because it is a costly Endeavor, but the fact that it
17:43
already exists, makes it less costly.
17:45
You don't have to go create it, and it's a diverse build thing.
17:48
And, you know, they have lots of tncs things to consider.
17:52
But yes, you are being spied on, but it's so I never really did
17:56
address, why, right? So it's not just to learn and
18:00
evolve their capability if they're smart, which I know at
18:04
least one of the large Voicebank holders is, they are turning you
18:09
into a predictive marketing machine, right?
18:13
So that's how they make their money is through advertisers
18:17
services. And Licensing is never where the
18:19
money's at. It's always in the services
18:21
associated with it. So they make Bank cold hard
18:26
money from the ability to predict what Jeff Steadman is
18:31
going to buy after he purchases that tent?
18:35
Is he going to be looking? For a camper, is he going to be
18:37
looking for a place in Arizona, right?
18:40
It's an amalgamation of what you're buying and how are you
18:45
searching? And what are you all totally
18:47
doing with it? Are you a giver meeting
18:50
Charities? What do we serve you to make our
18:54
platform? The most valuable to our
18:58
advertisers as we possibly can? So do they want to know that you
19:03
know you yelled at your wife and no, they don't.
19:06
Are that, although, that audio is absolutely there.
19:09
You can go here it on all platforms.
19:12
You can go and hear all of the audio that's been captured and
19:15
delete it, but that doesn't mean that it's being triple deleted
19:19
on there. And so it, once it's out there,
19:21
it's out there. But you can hear what's being
19:24
saved from your utterances. I feel like that this is where
19:28
the Ron Swanson death of him learning what a computer does
19:31
and then him just walking out to the dumpster and just throwing,
19:34
let's give her the doctor is where that might be.
19:37
That's very um, yeah, I mean, I think people, you know, if the
19:41
product is free, you're the product is kind of some kind of
19:43
way to look at it is, yeah, advertising, I'm sure I'm lumped
19:46
in with other folks, and it's just another key metric around.
19:50
Well, Jeff just by guitar and he's trying to learn how to
19:52
play. And, you know, if you are video
19:54
with me, you can see there's one of the background, I am I Suck.
19:57
So, but at some point, I'm probably gonna be looking for
20:00
lessons or maybe a guitar, sand or guitar picks, right?
20:03
And I'm in a certain age range. All of these are data points
20:06
that come together to kind of formulate, who is what is this
20:09
entity look like, and the voice Biometrics that I'm providing.
20:14
Whenever I talk to Siri Google, I'm sure you just activated a
20:17
bunch of stuff out there, right? Alexa, whatever may be all feeds
20:21
into that. What?
20:22
I find a little bit interesting is At this point, we've got Siri
20:27
at least in the consumer side of got Syrian, we've got Google, it
20:31
sounds like a Lexus. Having some issues because
20:33
Amazon is reading articles about how Alexa is not doing great and
20:37
they may shut that down or pair of back considerably, Microsoft
20:40
got out of the Cortana game a while back.
20:44
So I don't either winners and losers already in this space
20:47
when it comes to the voice side. Not really.
20:51
I think it's more the value prop ride.
20:53
So they're not seeing a Lot of Roi on the invoice, the
20:57
Investments they've made in their voice capabilities.
21:00
Primarily because in my never humble opinion, they missed the
21:03
boat on why to do it to begin with, or if you're doing it
21:08
while you're doing it for its initial intended, purpose,
21:11
expand your purview beyond what you operationally decided to
21:16
invoke or that led you to invoke the use of Voice or the creation
21:21
of that capability and they miss the boat.
21:25
So I don't necessarily say it would say that there's a player.
21:29
It's the how, right? So you could it's like I would
21:33
liken it to you. Could find a bazillion, really
21:38
good C sharp developers but if they don't know architecture,
21:43
you're not going to want them to build to be some.
21:47
It's very, it's a really crappy analogy.
21:51
I'm trying to give a better one, but my point is that It's about
21:55
Who and the experiences of the deployer and there are lots of
22:01
algorithms out there and they have multiple purposes and they
22:04
serve better and different options to the point that
22:08
there's now multi algorithmic Fusion, where, you know, there's
22:13
these eight algorithms, that nail it for most of your
22:19
Universal Baseline population, and then you put them together.
22:23
And then you, you lesson, And lessen, your false, your
22:28
duration. It takes to authenticate your
22:30
duration to time is money, especially the call centers,
22:35
right? It's like what?
22:35
I think the industry standard is a dollar, something a minute,
22:38
that's insane. Your Roi is sucked up the
22:41
minute, you get a dumb customer, right?
22:43
Are you good? And I Karen.
22:45
I think there's none of those out there and, of course not.
22:49
But when you get those, their profitability in the cost to do
22:53
business is Trashed, right? So they have to look at all the
22:58
different scenarios where they could salvage or expedite or
23:02
minimize that unpredictable expense Associated.
23:06
So, let me pivots like slightly because I want to talk about
23:10
deep fakes. And how does that impact things
23:13
like voice Biometrics? And we've seen a lot of news
23:16
headlines around it, you know. I think the whole Jordan Peele
23:18
Obama saying, what you know, was was popular.
23:22
I don't know, it seems like 10 years ago.
23:23
It was probably like to How impactful are deep audio fakes
23:29
to the efficacy of trying to do, you know, strong identification
23:33
or authentication through voice biometric.
23:36
So right now that there's a lot of dependencies there, that's a
23:39
loaded question but to date it is absolutely avoidable as the
23:46
Technologies evolve for deep things, especially around audio.
23:51
It's going to be tough. So the industry needs to be
23:55
Typically Progressive in figuring out how to stay in the
24:01
proper place for the betterment of black hats and convert them.
24:08
That's all I can start converting them now so that, you
24:11
know, as this evolves or there because I would say, we're
24:15
probably at a, at most three to five years threshold to somebody
24:19
being able to have an epic deep fake maker.
24:24
But We're not going to be able to contend with, so that's kind
24:28
of why the Department of Defense and DHS are evolving Beyond a
24:32
single biometric. So while VB is absolute, there's
24:37
so many things you can do with it when you fuse it with other
24:40
biometric. Modalities your false positives
24:46
are almost completely eliminated.
24:49
I mean, it's measurably significant.
24:52
When you talk about Multimodal. But it's complicated you know
24:57
and if they're not if they're not Progressive enough and their
25:01
dedication to Biometrics it's a tough it's a tough pitch to
25:04
make. But once you dig in and you look
25:07
at the ROI compared to a single modality, it gets much much more
25:13
palatable of a conversation to have.
25:15
You've mentioned this concept of fusion a couple times.
25:19
Can you explain what that is to my simple brain?
25:22
No you're fine. So it's really each independent
25:27
biometric modality being fused together from a score point of
25:31
view. So I've taken a score say from
25:33
your voice, I'm taking a score from your facial.
25:36
I'm taking a score from your from an old factory biometric
25:41
and I'm fusing those together to say, either your Bad guy or
25:47
infusing them together to say this guy could walk through this
25:49
door without having to present a card without it, which can be
25:53
stolen without having to do the iris stance or obfuscated with
25:59
contact lenses or violate. You know, successfully penetrate
26:04
the security measure. It makes that nearly impossible
26:08
because you may be amazing fakin one, but you're never going to
26:12
be amazing at faking free. So it allows you to set the
26:15
threshold. For entry significantly, higher
26:18
thereby, always accompanying, creating a lot more structure
26:24
and security. You said olfactory biometric.
26:28
So as at the sense of smell. Yes, it is.
26:30
The sense of smell. Okay, this is news to me, I
26:32
guess, I'm not open my Biometrics game but I didn't
26:34
realize that that was something that was out there and I, I'm
26:37
intrigued very much how stinky a person is, no, no.
26:41
So or what is it, something else UT met.
26:44
So, it's used for me A lot of purposes, but the, the entree of
26:50
olfactory came from bombs and drugs.
26:56
So you have residual that you can't wash off, right?
27:00
And so having a dog isn't always possible, who's trained to
27:05
identify, and it's difficult for them to segment between like,
27:10
I'm if you followed Casey Anthony that trial where they
27:14
said the Smell of maggots produce the same smell as a
27:19
decaying carcass to a dog. So that doesn't happen and all
27:24
Factory Biometrics. And that's what?
27:27
What drove the introduction of olfactory Biometrics, but it's
27:31
super cool because then when you get nervous, you actually emit a
27:34
smell, it's a heat sink. There's a heat sink pattern as
27:38
well that we can look at. So, when we put all that
27:41
together, we know if you're a nervous wreck, which means you
27:44
probably did something wrong. Or if you're fixing to come
27:48
shoot somebody because you've got a bomb hidden in a backpack,
27:52
you know. So, olfactory has become pretty
27:55
sweet with identifying bag. There's probably not a whole lot
28:00
and Industry were that's applicable.
28:02
I've been like the airports train stations, you know,
28:05
Transportation modalities. But yeah, it's super cool,
28:09
right? How important is the quality of
28:12
the audio with the identity of the sender?
28:16
Strive for excellent audio. Yeah but you know, when you're
28:19
here at the pregame show pregame was it's always many hurdles to
28:25
overcome. But how important is it that the
28:28
quality of the input or the recording is at a high level, to
28:34
avoid false positives? Or is that just a factor if
28:38
that's a factor you have to contend with it, right?
28:40
But it has evolved so far and doing sound separation.
28:45
And if you're have any familiarity with natural
28:47
language processing and how far that has come, that's it's
28:52
giving us a very targeted method of separating sound from voice
28:59
but in the cases of identifying fraud or Bad actors in certain
29:03
scenarios obviously in the dod, it's important to hear what's
29:08
around as much as it is to hear what they're saying, how they're
29:11
saying it. I'm so quality.
29:15
So the give Factor there is I need more audio if it's really
29:21
bad quality, as opposed to if you're sitting in this scenario
29:26
where our audio is on fleet, that's right.
29:30
I can capture you in literally a second half of audio.
29:35
But yeah, so that's where that is.
29:36
Now, if you're recording and it's like three seconds in a
29:40
subway, station is probably, maybe not enough.
29:44
Not for one time if you are in and out of different places,
29:50
like when you call your bank and want to use the audio
29:54
authentication or the voice authentication, if you've had a
29:59
cold before and if and of called into the bank, if you have
30:03
called from a Subway, if you've called from the airport standing
30:08
underneath, you know, the allowed jet engine it, learns
30:14
right. That's the machine learning
30:15
elements and beauty that is that lays in voice biometric modern,
30:19
voice Biometrics. So yeah, there's a lot of
30:23
factors there. Ma'am.
30:24
I could see you as an expert witness in a trial where you had
30:29
like three seconds of audio recorded in a subway station and
30:35
then you start running through the scientific like all right
30:39
we're 78 percent sure based on this and 48 percent.
30:44
Sure. Based on this.
30:45
And therefore, we have a 99% shortness that this is the same
30:50
person. Absolutely.
30:52
So. Okay, well I think we'd beat
30:55
that one, pretty good but I wanted to bring up kind of in
30:58
the I am space where I've run into voice Biometrics a few
31:02
times as like, hmm. I wonder if this would be a good
31:06
application. I'd like to get your opinion so
31:08
it's what we call, quote, unquote, the cleanroom scenario.
31:12
So it's the idea that we want to Do MFA, but the person who's in
31:18
the clean room, whether it's, you know, the typical like
31:22
situation where somebody's wearing like a bunny suit or,
31:26
you know, potentially it's like in our prison.
31:28
But anyway the person can't bring in their phone so you
31:32
can't send them a second Factor. The question is, is that a good
31:38
scenario to use voice because my understanding was is that like
31:44
with a lot of Biometrics you have device specificity right so
31:50
you have to go through the registration on each device.
31:52
But with boys the voice is being stored in kind of a central
31:57
location or the voice fingerprint, which is probably
32:02
just the number along, no long strand.
32:05
A am I thinking about this right?
32:08
You're absolutely right. And if you're talking about like
32:11
it, especially if you look at jail or prisons, it's a hell of
32:15
a use case, right? Because you voices absolutely
32:19
there. Everybody talks about what you
32:21
gonna do when you get a deaf person.
32:22
So I mean not that that happens all the time but there is
32:30
something to be said about that being a perfect case for Fusion
32:34
because not only Will you identify you'll be able to tell
32:37
if they have bad intentions. So if somebody's coming in and
32:40
they're hiding something, you know, even the best people
32:44
readers there's always someone who's going to be better more
32:48
Charming or more effective at getting something in.
32:51
That's bad, it's super cool. Use case for biometric Fusion.
32:58
So I think that that kind of Central Storage of the
33:03
fingerprint of The Voice, Just kind of one of the advantages
33:08
are differentiators between VB and other forms of Biometrics,
33:13
but I think the other thing is, you know, in a lot of those
33:16
scenarios. If somebody's wearing a bunny
33:18
suit right there, they're not going to be able to do a
33:22
fingerprint, they're not going to be able to necessarily do
33:26
facial recognition, things like that.
33:28
So it might be one of the few scenarios where boys are voice.
33:32
Might be one of the few options. And I guess what I'd like to
33:36
also ask is, you know, are there other use cases that you see
33:41
voice? Use them a lot for
33:43
authentication? Where it makes sense other than
33:47
fin serve well. So, so good.
33:49
Tell me what you mean, by that Financial Services.
33:53
No, I'm sorry, I need that part. What do you mean in terms of
33:56
using voice for Finn serve as that?
33:59
Just like when you call the call center sometimes but like Bank
34:03
of America, for example. As doing, I don't know if they
34:07
have rolled it out, nationally or globally yet, but they've
34:11
created banking centers that don't have people.
34:14
So you can literally walk up to a screen that is about below
34:19
your waist to probably depending on your height.
34:21
Of course. For me, it's like a foot over my
34:23
head. So I would say it's probably a
34:25
four or five foot tall screen with AI, it has an artificial
34:30
agent, on the other side, and I don't have to authenticate, I
34:34
exist. So that I never saw some in the
34:37
near, the subway stations on Broadway in New York City.
34:41
There's a couple in Columbus, we're such a test a test Market,
34:45
it's silly. We get the first flavor of lays
34:48
when they first release it to see how we do, which can be good
34:52
or bad depending on all right. Exactly.
34:55
Cinnamon Toast Crunch flavored Doritos.
34:56
No, thank you. Anyways, so they they the use
35:02
case there is wicked cool because You literally don't have
35:07
to do anything but walk in and conduct your business.
35:11
It's a sweet use case if you think about applying that theory
35:16
to something like Walmart think of how Walmart could use that
35:20
from a predictive marketing point of view.
35:22
So they follow a pattern of life, right?
35:24
Or pattern of your existence throughout a store.
35:27
They have security cameras anyway.
35:29
So use that data, right? So, watch where they're going
35:33
and what are they buying? And then, Order.
35:36
And so what would they like Leah by an Impulse buy at the
35:40
register? Or you noticed when these
35:43
products are close to these products and this one gets
35:47
purchased because it's right behind them and increased sales
35:50
of these items. So it's we got to stop thinking
35:54
that it's just about authentication.
35:56
It's about making I'm a capitalist.
36:01
Can you tell? It's making money and it's
36:04
making me. Money more effectively, which
36:06
would drive costs down. I think the economic
36:10
ramifications are pretty big. If we get out of our own way and
36:13
stop worrying about the false connotation of Biometrics, being
36:20
an invasion of privacy, because it's not, we're not holding your
36:23
voice. We're retaining your audio.
36:26
We're triple hashing. Creating a big long.
36:30
Number that if you stole it, even if I send it to you, you
36:34
would not be able. First identify, even that it's
36:36
yours. All right, so there is no pii
36:40
involved in. VB it just It Isn't.
36:45
So once we get out of that and stop worrying about that, then
36:48
we could talk about what can really be done with by my church
36:51
soapbox off. So what about scenarios where
36:55
you definitely do not want to use like the opposite, right?
36:58
We talked about a couple use cases here where yeah makes
37:01
sense, you know, prison or a jail or financial services or Or
37:05
retail in general. We think being able to walk up
37:07
and say, hey, I want to know about a quarter pounder with
37:11
cheese. Only onions and fries and a
37:13
coke, right? Rather than talking to a person,
37:15
which appears to be going by the wayside, or going after those
37:19
one stupid little iPads and keying in your own order, right?
37:22
The might actually have the order, right?
37:24
So I'm good right? Yeah, I'm okay with that part 2.
37:27
But what's the opposite of this? Like, where did where is voice
37:31
Biometrics never to be used? I don't think it exists.
37:37
There's no scenario where it just doesn't make sense, to
37:39
apply voice Biometrics as part of the authentication string or
37:43
the identification not for me. Even I like a loud loud.
37:47
Place quiet place place maybe with a lot of voices like an
37:52
airport or a subway station. No isolation false, anything
37:56
like that we already do it, right?
37:59
There's a there's an effort that was done by the Air Force and
38:03
and the Department of Defense is SNT program which is their
38:06
science and Technology. It's kind of like the evaluation
38:08
space for the Department of Defense where they literally can
38:13
authenticate a pilot standing at his aircraft while it's running
38:18
the they can do all kinds of Funky Cold Medina stuff.
38:23
Now that I can't think of a single scenario where it would
38:27
not be a relevant possibility, okay.
38:32
How about it might be capable but what point does the cost?
38:37
The benefit are these scenarios where it's super expensive to
38:40
have, you know? Okay.
38:42
Well do we really need to be able to identify a pilot outside
38:45
of his F-16 while the engines are running?
38:47
Yes, you can imagine if that's cheap to do not see.
38:51
It's well that would not be on the heavier side of cost.
38:57
It's really what what drives cost and biotin voice.
39:00
Biometrics speed speed to authenticate the thresholds
39:05
you're willing to accept. Depth.
39:07
And then that's about it. That's probably it speed and
39:13
accuracy or those metrics that a buyer of software can really
39:21
gauge or get that information from a software vendor.
39:26
So if I'm in the market for a VB solution, are those the things
39:31
that I should be looking at and are they metrics that the vendor
39:36
will Provide and I can rely on their answers.
39:40
Okay, well I got me until you threw in the end telling the
39:44
truth part. Yeah.
39:46
So I mean there's a lot of variation there too.
39:48
It's kind of like I feel like I'm not giving you any straight
39:51
answers but it's true. I mean, when you look at, if
39:55
you're in the market for VB solution, you need to solidly
39:58
understand why and what are your requirements meaning like, do I
40:04
have an alternative flow that It's less expensive if I give up
40:09
but a percentage or if I require a little slightly longer
40:13
elongation of audio acquisition. You got to look at all that
40:18
before you have the conversations of percentages.
40:21
But like in a, in a financial entity, you're not going to want
40:25
anything less than 99.9% by even would prefer a hundred, but 100
40:31
isn't gonna happen, but if I can acquire a positive ID, A hundred
40:38
and a hundred percent confidence in three seconds.
40:40
I'm n. But if it's 30 seconds, which I
40:43
know some banks are still on that old Antiquated technology
40:46
where I got to run my mouth and tell you 5 times.
40:49
How much I love my to use my voice as my password.
40:52
No, thanks. Not going to happen.
40:54
I'm just gonna hit zero like a vicious lie right that.
40:58
No, that's clearly annoying but I guess I want a clarification
41:03
on that. So is it 30 seconds if I know Oh
41:07
okay. It's claiming to be Mel dial and
41:11
so I just have to check this boys against what I know for Mel
41:15
or is it? You are one of my 200 million
41:18
customers and now I have to use your voice to figure out, which
41:22
of those two hundred million customers.
41:23
Are you therein lies? The rub gym?
41:25
That is absolutely. The differentiator
41:28
algorithmically is inclusionary logic versus exclusionary logic
41:33
and the voice Biometrics algorithmic approach, It sounds
41:39
like it wouldn't make any difference but when you look at
41:42
it scientifically, the data is there that if I already know you
41:45
exist, like if there was a global voice biometric database
41:50
of everyone to authenticate anything then it's going to be
41:57
100%. Inclusionary logic meaning can I
42:02
I don't even look to see if you're excluded from the data
42:04
set. Exclusionary logic is Is where
42:07
they look at it and say. So can I exclude her from being
42:13
any other entity than Mel demon and so that algorithmically and
42:18
how you forge? The process is where that
42:22
becomes a differentiator so what's next for voice
42:25
Biometrics? Like what's the cool thing
42:28
coming up that we need to be aware of so it's here but
42:31
sentiment and intention are wicked.
42:37
And now it's here but it's not widely used yet.
42:42
I think it's worth looking at. If you're in an industry, there
42:47
are so many use cases. Where sentiment and phraseology
42:52
is really under estimated right now.
42:57
There's a lot of value there even if it's good versus bad or
43:02
if it's there really irritated or they say they're irritated,
43:06
they're really not or They make a false claim.
43:09
It's not like a lie detector but if you're multimodal fusing you
43:14
absolutely have a lie detector capability and multimodal fused
43:18
Biometrics. So it depends on your industry
43:22
of course. But sentiment phraseology is
43:25
wicked cool. And that is the factor.
43:29
Allowing that deception versus well-intended in the Also, the
43:37
biometric I've kind of seen that that sentiment in some recent
43:42
Sci-Fi films, and TV shows that it watching.
43:44
It's like, you know, you have the scanners, I call the
43:46
person's lying, just by the way, they're talking, how do you
43:48
establish sentiment? Like how does a normal person do
43:52
this? Or is it just the massive amount
43:54
of data that's out there and correlating it across the other
43:57
modes? That is not my area of
43:59
expertise. There's a couple wicked cool
44:03
scientists who are doing nothing but that and that the only
44:08
company I know right now who's really nailing it is next video
44:12
which was bought by nice systems.
44:16
They're the only company I've seen that has truly made some
44:19
strides there and how do you send him outside of the
44:23
government space? It's used all over and
44:26
government but for different purposes but the industry talk
44:32
to the scientists over a next idea.
44:35
Okay. I think it's probably a show
44:37
topic. Maybe 2023 given where we're at
44:39
right now in the air. Yeah.
44:40
I think I'm deaf agreed that one out.
44:43
What about AI when it comes to voice Biometrics?
44:47
And the use cases, I'm thinking here is data from Star Trek, a
44:51
walking talking Android that is a, you know, symbol chroma of a
44:55
human being. Do you think at some point that
44:57
they would have their own vows biometric?
45:00
Or would it be maybe something else behind this?
45:02
Seems like micro service API calls based on Bluetooth or
45:06
Wi-Fi. Signals that the robot is
45:08
generating or things like that, I would.
45:11
So I would say the capabilities exist today.
45:14
I know that there's a couple a eyes that are fully functional
45:18
and can do voice activated things.
45:24
But I guess the question is should they?
45:25
And how far do you want to take that?
45:27
But how that happens, it's going to evolve but at this point I
45:32
would say it can be on board. Absolutely, you can have on
45:34
board until robots, take over the world and we have to all
45:37
Will Smith mute so there's I don't know.
45:41
Remember the name of it there's it's all over the place right
45:44
now where it mostly actually in them and asia-pac region where
45:50
they're generating whole songs. Whole artists 100% with AI.
45:56
But I'm too old to remember the name and this is full-on
45:59
original. I guess.
46:02
Scores or songs or interesting upgrade.
46:05
Yeah, I've seen them do. I've heard my daughter's really
46:07
into this and I've heard EDM, I've heard an opera singer.
46:14
The who wasn't really officer was an AI.
46:17
I've heard, what's that? You said have ever leader Jim.
46:24
What's the name of that genre of music?
46:26
Are they country top? No, I didn't hear country.
46:31
I forget. But it's like an evolved gen Z
46:35
genre of music. But I forgive you yet, so it's
46:39
super cool. It's there and it's only going
46:41
to get better. This is that is the Fidelity of
46:44
that audio there or is it still mechanical sounding?
46:48
Oh, no. Not even a little bit
46:50
interesting. So I think when the yeah,
46:53
gotcha, well I think when, you know, You can take data points
46:58
and big data and crunch it all together.
47:00
I mean, they already know like beats for a minute what people
47:03
like and what, you know, how to construct songs, so that people
47:08
are attracted to them. Now, take that and start taking
47:11
what kind of voices people are tend to like and you mix all
47:15
that together and hit the themes that people like and now all of
47:19
a sudden you just have an algorithm algorithmically you
47:24
know made to be popular. Or song or even just customized
47:28
right to a person's flavor profile.
47:30
Yeah, now when I like our EDM Jim likes this country, right?
47:33
We're going to have different opinions over everything at
47:35
different point. Why you are our own unique
47:38
listening sound pattern, right? We could essentially generate
47:43
audio on the Fly that matches with our mood or whatever our
47:48
preferences. Our I think that is fascinating.
47:50
It is fascinating, it's like Spotify unwrapped on steroids
47:54
basically. Yeah.
47:57
All right, you we've been talking awful lot here.
47:59
I want to close out on a lighter note and got a couple options.
48:03
And the one that I'm going to pick here, is you have your own
48:06
late-night talk show who do you invite as your first guest
48:10
melting? Then it's the melting man show
48:12
hearing, I don't know every night at end on NBC.
48:16
Who's your first guest on your new show?
48:18
Do they have to be alive? No II didn't specify so you can
48:21
pick. Anybody want really, Jeff?
48:23
That's hard. I would have like a hundred and
48:25
twenty people. Pull off the bat that I would
48:27
say. I want to pick one or maybe pick
48:30
a wine up two or three guests. All right.
48:34
I would want Harriet Tubman because she was a badass, I
48:41
would want George Washington. And then modern day, I would
48:48
want. The creators of South Park,
48:53
Matt. And was that Matt Striker?
48:55
Interesting. Yeah, okay.
48:57
That's a lot years. That's a strong opening night,
49:01
right? I didn't think of historical
49:04
figures. I my mind went right to Elon
49:06
Musk. Oh, he's I have such a
49:09
girl-crush on him. Oh, be still my heart.
49:13
Yeah, that's that's my choice. What about you Jeff?
49:18
Yeah, I'm torn. So I'm going to pick two and
49:20
cheat. That's why I gave Mel cover to
49:23
pick whatever she wanted. I'm going with George Carlin.
49:28
As the opening guess and then Jon Stewart is sort of on the
49:32
bench there as well. Hurry up, that's who I would
49:35
start with why I'm a fan of comedy, and I think it would be
49:38
just a fascinating conversation, given how popular Carlin has
49:43
been for decades, especially even him and he passed away, you
49:46
know, few years back. But he's still such an influence
49:49
for a lot of comics and a lot of the genres of Comedy that are
49:52
out there now, feed off of sort of that, you know, not
49:57
necessarily Political but socio, societal kind of observations
50:02
and things like that. And I think Jon Stewart took it
50:05
to more of the political level of it, which I thought was just
50:09
hilarious. And I can, I could I could watch
50:12
The Daily Show from from years and years on end.
50:14
And Trevor Noah, you know, was his last show was last night, so
50:18
we were watching that one last night, but I would go with
50:20
George Carlin and Jon Stewart as the two that I would have on
50:23
there and then somebody historical.
50:25
I think would be cool to have like Abraham Lincoln.
50:27
You know what movie were you gonna go see things like that.
50:32
What movie? Yeah, movies back then I've got
50:37
a, I've got a share, my George Harlem story.
50:40
So yeah, one of my co-workers 15 years ago or something this guy
50:47
was hilarious and name was Patrick Gibbons and his father
50:51
was a drill sergeant in the Marines.
50:54
So you think like his dad must have Have been some kind of
50:58
like, you know, jerk like really, like overbearing and I
51:02
just couldn't imagine because this guy was just so funny.
51:05
Well, turns out his father, the drill sergeant in the Marines,
51:08
when he was a kid was best friends with George Carlin.
51:12
Nah. Yes.
51:13
As friends that grew up together and some guy like everything,
51:16
mr. George Carlin here.
51:18
Yeah. You know, I guess six degrees.
51:20
Exactly. I don't have the anti three
51:22
degrees now, right? Yeah, because I'm Jimmy, right,
51:25
we're both connected. Yeah.
51:27
We're not hit the same rank my all right.
51:30
One of my Idols. So I think I think we'll go
51:32
ahead and leave it there for this week Mel you been really
51:34
generous to the time I want to ask for two more minutes.
51:36
What is something that people should take away from this
51:39
conversation about voice Biometrics?
51:41
If anything going in wrapping up 22 going into 2023, what do I
51:48
need to know? Biometrics is not a privacy
51:51
concern. It's not pii.
51:55
And number two, use it. There's a use case for it
51:58
everywhere, the end. Okay.
52:01
Do you need two minutes Jimbo yourself?
52:03
Final thoughts for this week? Yeah.
52:04
My final thoughts are, I reached out to Mel on a, you know, I saw
52:09
her connected to Becky archambault at Gartner.
52:13
It didn't know Mel from, you know, anybody else or reached
52:16
out super awesome person, Person.
52:20
I think you reach out to her and make a connection.
52:24
But also, I mean, I would encourage you to take a chance
52:27
and anybody who's out there who you think you'd like to, you
52:32
know, make a connection with not to try and sell them something
52:36
please. But just to, you know, spark
52:39
that conversation or be connected and follow them or
52:42
whatever, go for it. Well, the worst they can do is
52:46
blow you off or black, you disconnect?
52:50
Yeah, that hurt over it. Yeah, there you go.
52:54
No, I think, you know, I think it's okay to ask for help right.
52:57
There are the problems that a lot of us face especially in any
53:00
space have probably been solved before.
53:03
Right. And then the stronger, your
53:04
network is the more people, you know, and more people that you,
53:07
that, you know, and that they know can help you get to that
53:11
answer quicker, right? We don't need to solve
53:12
everything for the first time, so yeah, ask for help.
53:15
I think that's ask for advice, right?
53:17
See how other people are doing? That's why I'm a big fan of
53:20
organizations like ID Pro, for example, and the slack Channel
53:23
they have, right? And you know, this podcast will
53:25
bring all kinds of viewpoints, you know, to, to the conscious
53:29
level. Hopefully, if you're not already
53:31
subconscious after listening. So, that's, I think, you know,
53:35
be aware of that. And I think Mel if you're cool
53:37
with it, we typically will put our guests LinkedIn connection
53:40
into our show notes. And if they have questions, they
53:42
can reach out and you know, maybe tell you that you're
53:45
wrong, or maybe that you're right, or maybe somewhere in
53:47
between, right. I've been wrong.
53:50
It's rare but it happens go to you.
53:52
If you tell Mel she's wrong, expect a response.
53:56
Probably a scientific way. Over analysed one.
53:59
Okay we're gonna go ahead and leave it for this week.
54:02
Want to thank everyone for listening and you can find us on
54:05
the web by Danny at the center.com.
54:07
We're on Twitter so far still at idac podcast and yeah with that
54:12
we'll go ahead and leave it for this week mellow thanks you so
54:14
much for taking the time. Jim as always thanks for the
54:17
time and we'll talk with everyone.
54:19
One in the next one. Thanks for listening to the
54:25
podcast. If you like what you heard,
54:27
don't forget to subscribe and visit us on the web and identity
54:30
at the center.com.