Latest / AI for Founders with Ryan Estes / Early warning for neurology and mental health with voice AI

Transcript

Early warning for neurology and mental health with voice AI · AI for Founders with Ryan Estes

0:00
Welcome back to AI for Founders. I'm Ryan Estes.
0:02
Today's guest is Henry O'Connell, the Co founder and
0:05
CEO of Canary Speech, the company turning your voice into
0:09
next generation health diagnostic.
0:11
This isn't sci-fi. Canary analyzes 2590 vocal
0:16
biomarkers every 10 milliseconds producing over 16,000,000 data
0:21
points per minute. And from that, it can detect
0:23
signs of anxiety, depression, cognitive decline, Parkinson's,
0:28
even multiple sclerosis just by listening to a conversation.
0:31
In this episode, you'll learn how Canary is transforming
0:34
healthcare through ambient AI listening, why your voice may be
0:37
the richest biometric after your DNA, and what it means when
0:41
doctors, algorithms, and empathy all work side by side.
0:45
This conversation is for founders building in AI or human
0:49
computer interaction, and for anyone obsessed where medicine
0:52
meets machine learning. Now, if you love big ideas with
0:55
a pulse, join our newsletter at AI for founders.co.
0:59
Each week we unpack how founders are turning AI breakthroughs
1:02
into billion dollar companies and cultural movements.
1:05
Don't just read about the future.
1:07
Subscribe and start building it. This episode is brought to you
1:10
by Kick Caster, the podcast booking agency that helps
1:13
founders become thought leaders. If you're ready to Share your
1:16
story on Top podcast, expand your audience and close high
1:20
intent deals through authentic conversations, Kick Caster makes
1:23
it effortless. We book entrepreneurs on
1:26
hundreds of industry shows every month, handling the research,
1:29
the outreach and the scheduling. So all you have to do is show up
1:33
and speak your truth. Turn your next podcast
1:36
appearance into your next big opportunity.
1:38
Visit kickcaster.com to apply and get started today.
1:42
That's Kickcaster by Moberst. All right, all right, coffee is
1:46
great, but coffee won't prep your meetings for you.
1:49
It'll just keep you awake while you panic, hastily scrolling
1:52
through LinkedIn at 6:59 AM. Ambient, however, will prep your
1:57
meetings. Each morning you get a founder
1:59
tuned briefing, sourced overnight from your calendar,
2:02
your meeting history, and the web.
2:04
It's like walking to every call with a CHEAT SHEET that you
2:06
didn't have to write. No more scrambling, no more
2:09
guessing. Just context, confidence, and
2:12
focus. And if you're raising money,
2:14
ambient investor dossiers are an absolute must have.
2:18
You'll know the investor's portfolio, their interests, and
2:22
their latest moves before you even step into the room.
2:24
You'll even get a heads up on red flags.
2:27
Think of it as the caffeine upgrade that your calendar
2:29
actually needs. Prep while you sleep, perform
2:33
while you're awake. Sign up for free at ambient dot
2:36
US. Go check it out at ambient dot
2:38
US. That's ambient dot US.
2:56
All right. We're ready to go, Henry.
2:58
Thanks for joining me today. Well, it's wonderful to be here,
3:02
Ryan. Thank you very much.
3:04
Absolutely. I've been very spoiled that I've
3:06
had been having a lot of conversations lately about the
3:10
intersection of AI and the the medical industry.
3:14
So I would love to hear briefly an introduction to Canary
3:19
Speech. Thank you.
3:23
So Canary Speech uses voice and conversational speech, ambient
3:31
listening to analyze for vocal biomarkers, 25190 of them
3:38
discrete data elements every 10 milliseconds.
3:42
And we have a that creates a data set of about 15 and a half,
3:47
16,000,000 data elements every minute.
3:51
Our voices the most complex motor function.
3:56
The creation of speech is the most complex motor function we
3:59
have, and it's directly connected with our central
4:03
nervous system, self consciously controlled.
4:08
So the central nervous system is driving the vocal cords and the
4:12
rest, the power and rate of respiration and our tongue
4:16
movement and our cheeks and everything else that just
4:19
creates speech. And at that level, everything's
4:25
identical, even though there are hundreds of different languages
4:29
around the world. So but all of the elements of
4:34
the articulatory system are the same regardless of what language
4:39
you're creating. And if there's a deficit in a
4:43
disease, a common cold or a progressive neurological
4:47
disease, regardless of what they did, is it impacts on the
4:51
ability of the central nervous system to create language or it
4:56
leaves a fingerprint in in the process of creating language.
5:01
And it that's what Canary speech is reading.
5:04
So when you have a patient doctor conversation, Canary
5:09
Speech is using ambient listening.
5:13
It's non intrusive. We're just engaged in that
5:16
relationship in a high trust environment, similar to the data
5:20
security that the hospital has. We function at that level, so
5:24
we're listening to the doctor patient conversation and in real
5:28
time we're returning up to a dozen different diseases to a
5:34
handheld device, a phone or a tablet that that the clinician
5:38
has, which may also be doing double duty and maybe doing the
5:42
transcription services for them. The doctors and patient notes,
5:46
we're just integrated fully into that platform.
5:48
And the wonderful thing about it is because of ambient listening,
5:51
we're not intruding into that relationship.
5:53
We're simply listening to the conversation.
5:56
And it doesn't matter what they're talking about.
5:58
They could be talking about, you know, a grandson's soccer match,
6:01
or they could be talking about the last fishing trip either one
6:06
of them was on. We're able to then analyze the
6:09
voice of the patient and return scores for anxiety, depression,
6:13
cognitive health, behavioral health, a whole range of
6:15
different things. And that allows as a, as a
6:20
clinical decision support tool, it allows the doctor to engage
6:28
with the patient in a more informed way with objective
6:31
information, maybe even helping them formulate the next question
6:39
or make a referral more accurately.
6:41
So that's what Canary Speech is doing.
6:44
Incredible. So you talk about ambient
6:45
listening. Is this hardware?
6:47
Is this software? Is this an app on your phone?
6:50
What what's, how are you capturing the the conversations?
6:54
Though we we use all common audio capture devices, video
6:59
conferencing, smartphones and apps, smartphones or or tablets,
7:07
telephone systems. We're on the Samsung watch for
7:11
instance. And so a person at home could
7:13
actually do an assessment using the audio capabilities of the
7:17
Samsung launch. We could do it remotely or we
7:21
can do it in an office. So ambient listening is a non
7:24
intrusive way to monitor with permission, of course, to
7:30
monitor the conversation that's already ongoing.
7:33
So we don't have to ask the patient or the doctor to, to
7:37
take a moment and give us an audio sample.
7:40
We simply listen to the natural conversation they're having it.
7:44
What it ends up doing is rather than taking a patient and
7:47
putting them in a room and say take this test, which can be,
7:51
which can be nerve wracking anyway.
7:53
You just simply are capturing audio during the process of a
7:57
normal conversation like you and I are having.
8:00
And I could be monitoring myself and I could have my anxiety,
8:04
depression, mild cognitive impairment, Huntington's
8:07
Parkinson's disease, all analyzed for, for either the
8:12
level of anxiety I have or the presence or absence of mild
8:16
cognitive impairment. So all of that happens without
8:20
intruding on the relationship between the doctor and patient.
8:23
And that was really key. We wanted to provide them with
8:27
an objective tool that provided actionable, meaningful
8:31
information, but without intruding in their relationship.
8:36
So that's what ambient listening allows.
8:38
Unbelievable. This is so interesting.
8:40
So what? What are the markers in voice
8:43
for, let's just say, anxiety and depression?
8:47
So because we have 25190 different things we look at,
8:52
machine learning helps us work out, we, we, we do feature
8:55
extraction of all of that every 10 milliseconds of audio and
8:59
that produces this enormous file, 15 and a half million,
9:02
16,000,000 data elements. And then machine learning helps
9:07
us work out a correlation between that and that
9:11
individual, that specific individuals anxiety level and a
9:15
doctor's diagnosis. So we've built models working
9:19
with clinical teams that have tight associations, tight
9:23
correlations with the clinical assessment.
9:27
And then independently, just from the audio itself, we can
9:32
measure then the anxiety of a person that would correlate to a
9:39
diagnosis or a assessment they would get from a clinician.
9:44
So that that's how it's built. This algorithm that that is the
9:47
result of that model building can then be deployed across
9:54
populations. Well, it's interesting that in
9:58
the application, I'm really interested to hear because like
10:01
stress and anxiety and and depression, everybody's voice is
10:06
going to be blowing up for those to varying degrees.
10:09
For example, I might have a little anxiety doing this
10:11
podcast, but that's good. That's why I do it.
10:14
It's, I just call it butterflies, but I can see the,
10:17
the, the value really if you're looking for precursor kind of
10:20
indications from Alzheimer's or Parkinson's, like, you know,
10:24
stuff that might be harder for a clinician, maybe just because of
10:29
proximity, maybe just because of the frequency of their visits,
10:33
maybe because people get really excited and some of those
10:36
markers aren't there. So finding signal in everyday
10:39
kind of ambient conversations, maybe you're able to identify it
10:44
early, Is that kind of the idea behind it?
10:46
It is. I mean Canary speech comes from
10:49
Canary in the coal mine. So an early warning system
10:53
miners used to take a cage Canary in with them and if the
10:57
oxygen levels were were coming down and the air might not be
11:02
safe, the Canary would respond to that before the miner did.
11:06
Then they'd grab their Canary and they'd run out.
11:09
You know, so you are correct in illnesses that are difficult to
11:16
identify early and particularly at the primary care level,
11:18
because we're deploying with, with primary care physicians,
11:23
with paediatricians with, with individuals that that may not
11:27
see specialized diseases. You know, they might not,
11:31
they're not the neurologist or the psychiatrist team that are
11:35
are see hundreds and thousands of these patients are familiar
11:39
with them. So the primary care is seeing
11:42
someone twice a year on an annual checkup and Canary speech
11:46
can in as I said, non intrusively it can measure about
11:50
a dozen different diseases and just simply alert them to the
11:54
potential that a referral should be made.
11:56
That's what the clinical decision support component of
11:59
this is. So we provide them based on
12:02
vocal biomarkers. We're providing them with an
12:04
assessment that would indicate that mild cognitive impairment
12:08
or that Huntington's disease or Parkinson's or Ms. or anything
12:12
like that might in fact warrant the referral.
12:16
Now in talking to some of the doctors, one of the one of the
12:21
chief medical officers at one of the hospitals we're working with
12:24
said their deployment is, is done like they do everything
12:29
else, intelligently with thought.
12:31
And they said, you know, if we have a primary care Doctor Who
12:35
is watching after a patient who has recently had cancer
12:40
treatment, we we already expect them to have elevated anxiety
12:44
and depression. So the primary care doc is going
12:47
to maintain that relationship and manage the patient through
12:51
that post treatment, you know, rather than make an immediate
12:55
referral. So this is again information,
13:00
but it but it's meant among the array of information a doctor
13:04
has to provide them with an ability to make a decision on
13:10
objective information with respect to a referral.
13:13
OK, Yeah. So the human, the human in the
13:16
loop will always be the Doctor. Canary is going to give
13:18
recommendation based on the data we've detected XYZ.
13:21
This might inform your opinion one way or the other, but
13:24
ultimately this relies on the the Doctor in the loop.
13:28
Yeah, absolutely. And, and they're, they're
13:31
looking, our brains are wonderful, as you know, and
13:34
they're looking at, you know, all kinds of information from
13:38
their experience and their training, their relationship
13:41
that they have with this individual or the relationship
13:44
they have with with 100 other individuals that are similar or
13:48
presenting similar characteristics at the moment.
13:52
They take that all together. That expertise allows them to
13:55
make a decision about the care of that patient.
13:58
This is another one of those elements.
13:59
What's nice about it is that it's objective and it doesn't
14:03
matter whose hand it's in. It's going to give you the same
14:06
information from person to person to person to person.
14:10
So it's an objective piece of information that's reproducible
14:14
in a clinical setting. Does the information also travel
14:18
to the patient? It can, I mean, that's up to the
14:21
doctor, of course, just like anything.
14:24
The information that's gathered on a patient in a clinical
14:27
setting is generally information that's communicated to the
14:30
patient. You know, they have a freedom to
14:32
know that, they have a right to know that and they'll
14:34
communicate with them on that. We do go home with some
14:38
patients. Again, if a, if a, if a clinical
14:42
team decides that the patient's treatment can be augmented or
14:50
improved or valued added by going home with our product,
14:54
they can go home with our product and they can do
14:56
assessments in between treatments or in between
14:59
meetings in an office, they can be doing assessments.
15:02
They can also use our product to do remote assessments of
15:06
individuals or virtual meetings on, say, a Zoom or a Teledoc
15:12
platform, a Teams platform, and we're integrated into the
15:16
Microsoft Dragon Copilot platform as well.
15:20
So when a doctor is using the Dragon Copilot to do
15:23
transcription services, Canary Speech is part of that and can
15:28
provide them with the assessments that we've been
15:31
talking about in real time. Readable.
15:35
I love it. It's so interesting.
15:36
A lot of the conversations we have on this podcast is about
15:40
kind of the the human computer interface and the success of
15:44
ChatGPT and the rise of AI in the last couple years largely
15:48
was due to the fact that people are excited about using the chat
15:52
function. When you sit down and you you
15:54
tap what you need into the little box and it responds and
15:57
you're chatting back and forth like a text message.
16:00
But without a doubt, we're going to have all this other signal
16:03
that we'll be able to interface with AI With me personally, I
16:07
think chats boring. I don't like typing, I don't
16:09
want to talk to it really, but I'm always talking.
16:12
So this becomes a really interesting signal that it, once
16:16
we have kind of like the, the devices that are, are keeping
16:20
multiple tabs on us, we'll say, you know, it's listening to you,
16:24
it's looking at you, it's smelling your hormones, it's
16:26
feeling your blood pressure, It's, it's taking tabs kind of
16:29
consistently. This becomes a very important
16:32
biomarker as like, how are you communicating?
16:35
And I, I've never really heard of this before, but it makes so
16:39
much sense the way that you said, you know, you know, the,
16:41
actually the, the biologically, how moving our mouth and, and
16:46
making sounds, it's very complicated.
16:49
So of course it's going to play a huge role in cognitive health.
16:53
But I mean, the, the deeper we get.
16:55
And I'd love to maybe this is a good question for the future.
16:57
What other kind of like biological markers do you think
17:00
can be unlocked from speech? Now you and I know startups are
17:05
built in boardrooms. They're built in the wild during
17:08
3:00 AM. Bug fixes, botched launches, and
17:11
that one meeting where everything almost fall apart.
17:13
That's what Code Story is all about.
17:15
It's a podcast where tech founders breakdown the make or
17:18
break moments behind their products.
17:20
What went wrong, what went right, and what nearly wrecked
17:23
everything. Hosted by Noah Labhart, Code
17:26
Story goes past the highlight reel and into the honest stuff,
17:30
late pivots, brutal bugs, and somehow coming out alive with a
17:33
product people actually love. If you're in the build stage,
17:36
the scale stage, or the please let this feature work stage, I
17:40
can relate. This show will speak to your
17:41
soul. Listen to Code Story at
17:43
codestory.co or wherever you get your podcast.
17:46
Just search Code Story. It'll be right at the top.
17:49
Because great startups aren't born, they're forged one hard
17:52
moment at a time. Code story.co Hey everybody, you
17:57
know what's crazy? I've been sitting on a goldmine
17:59
this entire time and I didn't even know it.
18:01
My Gmail inbox? Literally thousands of warm
18:05
contacts who already know and trust me.
18:07
But here's the problem every founder faces.
18:09
How do you turn those relationships into revenue
18:11
without feeling like a sleazy salesperson?
18:14
That's exactly why I've been using Warmstar dot AI and
18:18
honestly, it's been a game changer for my business.
18:20
This AI platform connects your e-mail, analyzes all your
18:24
existing contacts, and helps you identify who and your network
18:27
might actually be interested in what you're doing.
18:30
Then it writes personalized outreach messages that don't
18:32
sound robotic or pushy. The best part?
18:35
You review every single message before it goes out.
18:38
No spray and pray nonsense. This is relationship based
18:41
selling done right. My listeners know I only
18:44
recommend tools I actually use, and Warm Start has literally
18:47
helped me reconnect with people I'd lost touch with and turn
18:50
those conversations into real business opportunities.
18:54
If you're a founder struggling with outreach, check out
18:56
warmstar dot AI and use code AIFF to get $50 off your first
19:02
month. Trust me, your existing network
19:04
is worth way more than you think.
19:06
Beach, you know the 25190 features that we look at today.
19:11
If you just talk about the vocal cords, which seem very natural.
19:15
The vocal cords vibrate controlled by the central
19:20
nervous system. They vibrate.
19:23
The speed at which they vibrate impacts on the sounds that we
19:27
make, but they also this, the change of that vibrational
19:32
characteristic, whether it's speeding up or slowing down,
19:35
also is indicative of what we're, what we're trying to
19:38
communicate. For instance, if we're trying to
19:40
communicate alarm, we, we, we see an acceleration of the vocal
19:47
cords to a much higher tonal element and then a drop off, you
19:51
know, and they come back pretty quickly.
19:56
But one of the examples I give someone is if I'm, if I'm, you
20:00
know, downtown and I hear a scream.
20:05
I, I know if that's a, a grandmother or a mother at a
20:09
child's soccer match versus someone who just was in an
20:13
accident and there's a difference.
20:15
And, and I don't know exactly how our brain has learned that,
20:19
but I know the difference. You know, 1 is, is an excitement
20:23
yell and the other one is, is a fear yell type of thing, You
20:28
know, an accident has occurred or whatever.
20:31
And, and so it is so subtle, you know, and then combine all of
20:37
that with how our respiration is coming into play and how, how
20:42
are all the rest of the articulatory system, you know,
20:45
including our tongue and our cheeks and our mouth and our jaw
20:48
and how we're moving things. All of that is all self
20:50
consciously controlled. But the central nervous system
20:54
is doing that. So the impact is that you have
20:58
this very, very dense and rich piece of information that can
21:04
be, is affected by a range of different things, whether as I
21:08
said, it's a, it's a common cold or it's, it happens to be a
21:12
progressive neurological disease, you know, like
21:14
Parkinson's or something. And so the wonderful thing we're
21:19
working on Ms. right now, multiple sclerosis.
21:22
And multiple sclerosis is difficult to diagnose and often
21:26
goes multiple years without a diagnosis.
21:30
And we know that because an individual's first diagnosis is
21:33
multiple lesions on their spinal cord and in their brain.
21:38
They've had, they may have had Ms. for 10 years.
21:43
We're looking at identifying that as early as 10 years before
21:46
that. I mean, so to really be able,
21:50
because there are there's medications, there's
21:52
therapeutics that can treat it and prevent that those lesions
21:55
from ever occurring. So those are the kind of simple
21:58
positive things I think can can happen.
22:02
AI is integrated into all of this.
22:04
So not only does that sound like AI, you know, using machine
22:09
learning and building algorithms and deploying those, but the
22:14
whole concept of ambient listening required the creation
22:19
and use of AI. So when we're listening to each
22:23
other, we're capable of separating those two voices.
22:27
We can identify who the patient is and who the doctor is.
22:31
Speaker identification. We know when we have enough
22:34
sample. We don't need to have continuous
22:37
sample. So we can stitch things together
22:39
on the fly. 5 seconds here, 12 seconds here.
22:42
We know when we have enough and we also know the quality of that
22:45
data. So if we get a section of data
22:47
when a cart was going by and the noise noise characteristics were
22:52
too high, we'll just wait for another 5 seconds.
22:56
So we can do that all automatically.
22:59
And then the rest of it is a is the processing of that data
23:02
which takes truthfully milliseconds of time and.
23:07
It, I think where it could be helpful down the road too, is
23:11
the problem with neurodegenerative disorders is
23:13
the patient doesn't notice, you know, so, so as if if you, let's
23:18
say Canary speech is just running in the background at all
23:20
times and it's, you can start to get a, some kind of warning
23:24
where it's like, hey, Ryan, you might be having a stroke buddy.
23:27
Like we should probably send a robot to pick you up and bring
23:30
you to the hospital, you know? It's interesting you say that.
23:33
I mean it's intuitive. We are introducing the latest
23:39
version of our ambient capability is Canary ambient
23:43
continuous monitoring. We're introducing that at HLTH
23:48
2025, which is the end of this month in Las Vegas.
23:51
So that's a brand new, a brand new capability.
23:54
Again, we we added some additional AI to it.
23:59
We recognize when someone is talking, there are certain
24:03
applications. If we're operating within a
24:05
room, it's let's say a patient's room, we can actually monitor
24:10
that room for a, for aggression level.
24:13
Nurses are being attacked daily, you know, by patients or by
24:17
family members. It's causing a tremendous loss
24:21
to the industry, not only financially, which is estimated
24:24
to be 20 billion a year, but they're losing 20% of their
24:28
first year nurses They're. Getting attacked?
24:31
I haven't heard that. Yeah, they're getting attacked
24:34
and it's happening. It's happening.
24:36
I'm talking to to healthcare institutions that say, you know,
24:40
we've got 17,000 nurses and we have two nurses attacked every
24:45
day, every day. Oh my gosh.
24:47
And So what Canary can do is we can take, we have an aggression
24:51
model and we can take that model.
24:53
And actually there are AI monitoring systems in the room
24:58
for watching the movement of patients, whether they fall or
25:01
don't when they get up, go to the bathroom, when a doctor
25:04
comes in or a nurse come in, do they clean their hands?
25:06
You know, compliance issues and patient movement issues and so
25:10
on and so forth. Well, Canary takes its algorithm
25:14
and it runs in on that computer and it wakes up when someone's
25:18
speaking. And we can measure aggression in
25:21
as many as five people in the room at a time.
25:24
So we've, we've created a green, yellow, red light response.
25:30
They said we're going to show this at, at, at HLTH.
25:34
So a green, yellow, red light response that shows up on a
25:38
handheld device that the nurse has anyway.
25:41
And she looks at it before going in the room.
25:43
If it's green, you know, you, you're walking in and it, but if
25:48
it's yellow or red, you're going in with someone, you're not
25:50
going in by yourself, you know, So that's the kind of thing that
25:54
I think AI and speech technology can provide.
25:57
And, and the combination of them is what's important because a
26:02
lot of what we're not only using AI to do what we do, but we're
26:05
using AI to facilitate and deliver what we do.
26:09
And, and it's the entire spectrum.
26:14
The other element. And, and I was speaking with you
26:16
about this, the other element that's, that's just a, one of
26:20
the tickets to the game is you've got to be operating at
26:24
the same data security, the same integrity that a hospital
26:29
operates at. So if you're a, if you're a, a
26:34
company that's providing AI and services like we do to a
26:39
healthcare institution, you've got to be able to say to them
26:42
and actually perform at the same level of data security and data
26:46
integrity that they're performing at which which Canary
26:50
does. Of course, we operate in a high
26:51
trust environment where ISO audited and certified.
26:55
So there's a whole range of things that as you're building
26:58
this out, you just need to put them all together.
27:00
And if you do, then you're a good partner in healthcare.
27:04
Yeah, 100%. You know, it's interesting
27:09
around a couple of levels on one side.
27:10
I'm curious about the other side of the coin, which is to say
27:14
that like, let's say I have a baseline and then, but perhaps
27:19
for for whatever reason, I have some kind of neurodegenerative
27:23
disorder that Canary is going to help me at least not diagnose.
27:29
But like recognize what about the other side?
27:32
If if I have, if I'm at normal and there's a downside, what
27:35
about the upside? What about non normative states
27:38
of consciousness? So I guess what I'm asking is,
27:41
is there a use? Is there a use case where Canary
27:45
Speech could see if I'm in some kind of like heightened state of
27:49
consciousness, if I'm in some kind of enlightened state
27:53
potentially? And what I'm talking about
27:56
specifically is there's a great spot in Denver called the Denver
27:59
Zen Center that has an amazing lineage.
28:02
And I like to go there and meditate and I love learning
28:04
about Zen because it is kind of like a a path of disciplic
28:09
succession, which means the teacher is with you and they
28:12
monitor you. There's a vibe check so they can
28:15
see through what you're saying, how you're saying it, what your
28:19
responses are to Co ONS. Let's say, for example, where
28:23
you are in your path of enlightenment's a heavy word,
28:25
but where are you at in your path?
28:27
Is there an application where Canary, it might be useful to
28:30
say like, wow, Ryan, you're, you're running at 12 right now.
28:34
It's you're doing really well, you know, kind of like treatment
28:38
for the well, something like that.
28:41
And I think the answer is yes. I think that what, what we find,
28:47
and you, you see this too, and you probably feel it in yourself
28:50
that after participating in such an exercise, there's, there's AI
28:57
don't know how to relate it, yes, but there's almost like a
29:02
gentleness in, in your voice. You know, there's a settling.
29:07
And if the, if we can hear that in the voice and we hear that
29:11
from people who they enter a room and they bring calmness
29:14
with them and it's in their demeanor, but it's also in their
29:18
voice and how they're communicating.
29:20
And if we can hear that and, and perceive that just like we
29:25
perceive anger, that's, that's just another thing that is
29:30
communicated in subtle ways in our voice.
29:33
And I often say to people, it has almost nothing to do with
29:36
the words we're speaking. And I'm not exactly sure how
29:41
these elements get communicated to our brain.
29:43
Our brain is, of course, a wonderful machine learning tool.
29:46
I mean, it's one of probably the best on planet Earth and.
29:49
So far. Notwithstanding the dolphins, I
29:53
don't want to take it away from and also octopus or octopi
29:57
probably really smart too. But you know, we've got this
30:01
great machine learning tool and, and people who who doctors,
30:05
clinicians, nurses, they do this all the time, you know, and
30:09
parents are doing it with their children constantly.
30:12
We do it with our parents, our older parents, and we're able to
30:16
understand that they may be saying one thing to us, but we
30:19
we know that the truth underlying it is they're
30:24
suffering at the moment. And so we don't say to them, oh,
30:27
you're suffering at the moment. Because we have compassion and
30:30
empathy because we do the same. We suffer at the moment
30:33
sometimes, but we join them in that somehow, you know, we
30:37
embrace and we go for a walk with them, whatever it is.
30:39
You know, I think that I think the answer is yes.
30:42
The simple answer would have been just simply yes.
30:44
But if you can measure one end of that spectrum, you're
30:49
measuring the other end of the spectrum too.
30:52
Yeah. And I think that's a wonderful,
30:54
like, biofeedback thing, you know?
30:56
Yeah, where could have continuous monitoring that would
30:59
allow you to have really true biofeedback.
31:02
And I apologize if for doing that to you.
31:07
I I just saw my little battery come on with a red light.
31:12
Oh yeah. And then you got to plug it in
31:15
for sure. No.
31:15
And that, you know, I think for clinicians, a lot of these kind
31:18
of like indicators would be really helpful because it's got
31:20
to be, you know, A cause of burnout that you're, you always
31:24
getting kind of the worst side of your patients and they're
31:26
attacking your nurses and it's, it gets really intense.
31:29
If you have something that's always kind of listening and it
31:31
can say, wow, Ryan, when you got out of that float tank boy, you
31:35
were like very relaxed and calm for like 3 days maybe.
31:39
That's really good thing for you.
31:40
You know, I mean, I, I, I am, I'm just fascinated with the
31:44
application of AI across the board with a kind of our, our
31:47
medical system, largely because one, it's really frustrating
31:51
because compliance is so high. There's so many different things
31:55
you have to, to overcome regulation wise to get products
31:58
to market. And then you also have the
32:00
culture of kind of medical practice as well.
32:03
I mean, just anecdotally myself, I, I had to fire my doctor
32:07
recently, you know, after a couple of years because it's
32:10
like you're just not giving me the kind of care that I'm
32:13
looking for. Now what I see where kind of
32:16
medicine is like moving to is, you know, the more of a combined
32:21
effort. Like I want to have informed AI
32:24
helping me through. I want to be doing telemedicine.
32:28
I want to be able to send a video that I can communicate
32:31
with my doctor. But these are things that
32:33
they're just unwilling to do. Like, oh, I, I don't want to do
32:36
a telehealth check and you have to come in.
32:38
It's like, well, boy, that takes 4 hours out of my day for a
32:41
routine checkup. If there's nothing wrong, you
32:44
know what I mean? Or I'd like to get my, my, my
32:46
blood drawn and done. The panel over here where it's
32:50
convenient to me and you can send the results over there.
32:52
And that becomes problematic largely just because people have
32:55
certain ways they like it done. So we're kind of at this place
32:59
now where we have these new therapies, whether it's peptides
33:03
or it's hormone replacement therapy.
33:05
There's a controversy around all these things, but like it's in
33:08
the culture, you know, where people are looking for
33:11
performance enhancing therapies, people are looking for AI aided
33:19
information. I know that this is also
33:21
controversial where they're like, OK, we'll provide the
33:23
information to the the physician, but we'll keep it
33:27
away from the the patient because largely the patients can
33:31
be like, hey, AI is recommending this and you're saying the
33:34
opposite based on your experience.
33:36
But now I'm questioning which is more valid, You know, so
33:39
there's, there's a lot of friction right here, which I
33:42
think creates kind of an environment for something really
33:44
cool to come out of it. What is your basic take on where
33:48
we're at as kind of like a healthcare industry and the the,
33:53
the potential for improvements in the some of the bottlenecks
33:57
that you're running into? Sure.
33:59
So right now I think we're where people are comfortable, it is a
34:06
clinical decision support where you're trying to create a force
34:10
multiplier really you're trying to provide more information to
34:14
help improve the quality of care and through accuracy and
34:19
objective meaningful information while they're still with the
34:22
patient. So it's not, I go take a test
34:24
over here and that information gets to me after the patient has
34:27
left and I look at it and I go, yeah, I, I don't have the time
34:31
to touch that. And they're coming back in six
34:34
months. You know, they just don't have
34:37
that time because they're seeing so many patients and the demand
34:41
is so high on them. So what we're trying to do is
34:45
provide these clinical assistance, clinical decision
34:52
support information, something that augments what they
34:55
currently have in real time, doesn't intrude in their life.
35:00
And when you, when you take that approach right now, and that's
35:03
where people are and, and AI will grow to, to be bigger and
35:09
the tools will get better and they'll get, they'll get more
35:15
expansive so that they'll be more information for immediately
35:19
for a doctor to see, you know, and a well trained Dr. will
35:22
embrace that information and, and try to translate that to the
35:26
care of their patients. And, and I think that's going to
35:29
take a period of time in the healthcare industry, maybe 5 to
35:33
10 years for full adoption of, of things were right now, as you
35:37
know, you can do remote, if I'm not overstating it, you can do
35:42
remote surgeries using robotic surgery capability, you know,
35:47
where you can have a trained surgeon in, in Massachusetts
35:52
working with a team, but a but a surgical instrument that's being
35:58
driven by him in Massachusetts because he's the expert at it
36:03
and he's able to do this remote operation on a patient.
36:08
And those things, those things have been around for a while.
36:11
They'll probably continue to expand.
36:13
I think they're going to be at the at the fingertips of, of
36:17
trained clinicians and doctors who have spent thousands of
36:20
years, thousands of hours of their life learning that trait,
36:24
you know, understanding it. So that when you go in, if
36:27
something is slightly different, they've seen slightly different
36:30
100 times or 1000 times, they're able to deal with slightly
36:33
different because they've seen it so many times.
36:36
Machines may not immediately recognize slightly different
36:40
until they learn it, but they'll learn it generally at the hands
36:44
of a clinical person. They're going to be more
36:47
intelligent because they're working side by side with a
36:49
clinical person. We get we gain we're capable of
36:54
building models across a whole range of different diseases, but
36:58
our ground truth is the diagnosis done by an expert in
37:01
the field and then we correlate to that.
37:04
Now what does that do? Well, it's standardizes that
37:07
assessment to to a team of experts and then that
37:13
standardization could be used as an objective piece of
37:16
information across 1000 different physicians.
37:19
You can even take it down to the primary care, allowing them to
37:23
make informed decisions about referrals to the specialist.
37:27
So the specialist is seeing more appropriately people they need
37:31
to see, you know, so a force multiplier in that case,
37:36
representative. Very cool.
37:38
OK. I'm going to just kind of read
37:40
out some of the research that I've done on the company and
37:44
once I'm done I want you to correct what I got wrong and
37:48
then fill in anything I missed. So we have a founding year 2016
37:55
raised about 13,000,009 issued patents, over 2500 speech
38:01
features and biomarkers are analyzed.
38:04
We've have roughly 40 employees about 8,000,000 per year and
38:10
estimated revenue. What am I missing about?
38:14
Can we? Well, we now have 14 issued
38:17
patents and 12 pending. We have raised a total of $26
38:23
million to date. The rest of it is pretty
38:27
accurate. This episode is brought to you
38:32
by Moberst, the AI first growth engine trusted by iconic
38:36
startups like Calm, Robin Hood, Reddit, Uber and Google.
38:41
Moberst is a digital, full service, mobile first marketing
38:44
agency that helps companies to scale and become category
38:47
leaders. Whether you're finding product
38:50
market fit or ready to accelerate growth, MO Burst
38:53
delivers creative, data-driven solutions that help with scale
38:57
and efficiency and impact. At MO Burst, we're not just
39:00
guessing what works, we're predicting it.
39:03
As an AI first agency, MO Burst uses cutting edge tools to
39:07
analyze user behavior, adapt creative in real time, and
39:11
uncover hidden growth opportunities.
39:13
The result? Marketing that's faster,
39:15
smarter, and built to convert from early stage startups to
39:18
global brands. We've helped hundreds of
39:21
companies overcome complex challenges, unlock new
39:24
audiences, and optimize every metric that matters.
39:27
At MO Burst, we believe that creative thinking, advanced
39:30
technology, and data are the core drivers of success at
39:34
scale. If you're serious about scaling
39:36
your business with precision and purpose, it's time to talk to MO
39:40
Burst. Visit moburst.com and let's
39:42
build something category defining together.
39:48
Hey founders, let me ask you this, are you building with AI
39:52
or getting left behind by it? If you're serious about scaling
39:56
smarter, faster and bolder, you need to check out AI for
40:00
founders.co. It's your go to resource for
40:04
cutting through the noise and actually using AI to grow your
40:07
startup. Each week, AI for Founders drops
40:10
a sharp, no fluff newsletter, dives deep with expert guests on
40:15
the podcast, and hosts monthly workshops where you can get your
40:18
hands dirty with the real tools and strategies.
40:22
Whether you're precede or Series B and beyond, launching
40:26
something new or just staying sharp, AI for Founders gives you
40:29
the edge you need to win in 2025 and beyond.
40:33
Hit AI for founders.co to subscribe to the newsletter,
40:37
tune into the podcast, and grab your spot in the next workshop.
40:41
And it's all free. Build smarter, move faster,
40:45
think bolder with AI for Founders.
40:48
So the patents. Talk to me about the patents and
40:50
and particularly the importance kind of a a health tech type
40:54
business. Sure.
40:56
So in eight, this is true of AIA, lot of AI is being driven
41:00
by open source software. So a lot of people, you'll see
41:05
200 companies in the US alone who are doing avatars for
41:10
patient engagement. There are no patents protecting
41:14
any of that. And so these 200 companies are
41:19
they're going to be, you know, five that emerge as the top
41:22
companies that that always happens.
41:24
And it will happen. They'll be more innovative,
41:27
they'll be more driven, they'll be better executors, meaning to
41:32
execute against plans and and produce the better mousetrap, if
41:37
you will. But none of it's protected by
41:39
patent. In the case of Canary speech,
41:42
when we looked at the field and I had the benefit of working
41:45
with a Co founder who has been a friend for 40 years, Jeff Adams.
41:49
Chef Adams led the team to build Dragon naturally speaking.
41:52
He built the first natural language processing for speech
41:58
to text. He led the team to build the
42:00
Amazon Echo. So I mean, this is a guy who
42:03
created new tech. All of those were brand new
42:05
technologies never existed before.
42:08
And in a bagel shop in 2016, we were talking about doing
42:12
something together and, and the healthcare field and speech were
42:17
a natural opportunity, but there had not been a successful
42:22
product in 25 years and part partly because their approach
42:27
was kind of wrong. And you spend 25 years doing
42:31
something, it's probably not right if you don't have a
42:34
product. And so we approached it from
42:37
vocal biomarker standpoint. So when we entered the field,
42:41
we, our first two patents created our first two families.
42:45
1 was the identification of bio or biomarkers in speech that
42:52
correlate with human condition and disease.
42:56
That was our first patent. So we look for features in
43:00
speech, not words, but features at the Bogle biomarker level.
43:04
The primary data area that correlate with diseases and the
43:09
ground truth would be a clinical assessment.
43:12
The second patent was the use of of vocal biomarkers to create
43:20
algorithms for the routine measurement of human condition
43:25
and disease in a clinical environment.
43:27
So those were the first two patents that we we built on and
43:31
we currently have those patents are issued in the US and Europe
43:35
and Asia and elsewhere. And in the US, each of those two
43:39
patents now have 4 patents in their family.
43:43
So then the next patent family was, we call it paired neural
43:48
Nets. But what it really accomplishes
43:50
is it says if you have met with an individual in January and
43:55
you're now meeting with them in June, we can actually compare
44:00
their vocal features, these biomarkers and give you a delta
44:05
on a range of diseases instantly.
44:07
We can tell you where they were six months ago, where they are
44:10
now and what the change is. And we can do that across time
44:15
longitudinally. The the, the next two patents
44:20
that we have applied for, they're pending now and those
44:25
patents have also issued in multiple jurisdictions.
44:28
The next two relate to this in this AI enhanced functionality.
44:34
So in order to make this practical, in order to our goal
44:39
was to put a tool in the hand of a, of a clinician or a clinical
44:43
team member that was practical. They didn't have to do jumping
44:47
jacks to make it work. And that's the ambient listening
44:51
and the AI intelligence, you know, speaker identification,
44:55
sample identification, sample quality, all of those things
44:58
come together using AI. So we call that AI enhanced
45:03
functionality for healthcare. And that's, that's a patent that
45:08
we've applied for. And then the third is to take
45:11
large language models which are not, are not discreet, meaning
45:17
they're, they're not, they're about specific individuals.
45:20
But when you gather a million of them, they no longer are, you
45:24
know, they're, they're general information about a condition.
45:28
And so that general information can give you an indication that
45:32
you might have a particular diagnosis or disease from that.
45:37
It's an indication. Yeah.
45:39
But it's not specifically about you.
45:41
What Canary speech does is listen to your specific voice
45:45
and say the likelihood that you have Parkinson's or Alzheimer's
45:49
or anxiety or depression is this so specifically you So, but
45:55
there is value in taking non specific information and very
46:00
specific information and combining them.
46:03
So the last patent area that and we've tested that it it enhances
46:08
the information. So the last patent area is the
46:11
fifth family of patents, is that one I.
46:16
Mean I just can't think of all these interesting applications.
46:20
You know, one, I'm a singer and so I I love any kind of
46:24
application of the voice, especially ones that are non
46:26
intuitive. This is kind of predictive,
46:29
generative work based on a relationship.
46:31
You would it would get to know your voice.
46:33
A couple of them come to mind. You know, let's say that the
46:37
Venn diagram of domestic violence and raised voices is
46:42
probably a concentric circle. You know, if you have somebody
46:47
who's out of jail on parole or something, you know, and you
46:50
have an app that could say that Alexa is starting to notice if
46:54
your voice is going up, you know, in volume, then it
46:58
probably, it can just be like, hey, perhaps we should call and
47:03
make a reservation for a massage.
47:06
Or maybe you should take a walk around the block there.
47:09
There's a lot of different things you can do with voice.
47:11
I'm thinking about myself, like I'm a pretty calm person,
47:13
generally speaking, but I'll get loud every once in a while.
47:17
And that is if my sports team is losing, you know, or lost, then
47:22
I'm going to be really upset. And it's confusing to the cat.
47:25
Very acceptable emotion, incidentally.
47:28
So I thought about, I was out walking, I walk every day and I
47:32
was out walking the other day and my, my watch vibrates when I
47:37
hit a mile and vibrates when I hit because I walk about 6 to 8
47:41
miles a day every day. And so, and then it gives me
47:45
what my, my average for that mile was and what my overall
47:49
average is and blah, blah, blah, you know, but it vibrates.
47:53
And that causes me to look at it.
47:56
Well, imagine we were using Canary ambient continuing
48:00
monitoring, continuous monitoring.
48:02
And when I got to the point that you were talking about, my watch
48:06
vibrated. So it just vibrates to say to
48:10
me, hey, you're getting in an upper level here.
48:15
You're staying. Yeah, and it could bring me and
48:17
I was thinking about this on this, this walk because it's
48:20
vibrating. And I'm going because I'm like
48:22
you. I wear all my emotions on my
48:24
sleeve. Yeah, every, every emotion.
48:26
People look at me and they go, So what are you upset about?
48:29
And I'll go. I'm not upset.
48:32
You know. OK, I am a little upset, but I
48:35
wasn't going to respond because I'm older now and I'm able to
48:38
control that, but I can't control my mouth.
48:40
You know, I mean, I can't control what my face is showing.
48:43
And but, you know, if if we were, if we're in a conversation
48:49
with someone in our watch vibrates, that could be like,
48:52
hey, that's personal, calm it down, you know?
48:57
Or I could be like, you know, like Henry Canary's telling me
49:00
you're yellow. Are we good here, buddy?
49:02
Like. What's it called exactly?
49:04
Yeah, that's the exact this kind of green, yellow, red thing.
49:07
Or if if my watch has noticed that over a period of days, my
49:12
anxiety is, you know, 20% higher than it normally runs, 20% above
49:20
where it is last month, it could vibrate and say, you know, check
49:25
your anxiety, you know, just automatically for me and these
49:29
are things that we could do today.
49:32
And Speaking of that, I mean right now creating agentic
49:36
solutions is like I'm doing it on a weekly basis, looking for
49:38
opportunity. Is there an API I can pull from
49:42
Canary to make it an app for Alexa so that when it senses me
49:46
say certain words? If I could just you know it or
49:49
decibels or anything I wanted to monitor monitor speech wise, I
49:52
could create apps around that and create automations.
49:56
So Canary again very intuitive, Canary has this platform of
50:01
capabilities and right now we're focused on the healthcare space
50:07
specifically on this doctor patient interaction, whether
50:10
it's in a video conference or it's in a real time meeting or
50:14
over a phone. However, these capabilities we
50:18
have taken those capabilities in order to support them.
50:21
We have, we have written interfaces that allow us allow a
50:27
person in customer support who doesn't have an engineering
50:30
degree, has a reasonable knowledge level, but not an
50:34
engineering degree to go in and actually design an experiment
50:38
for a customer. Customer calls and says, I was
50:40
trying this and it wasn't working.
50:41
And they go, let me, let me get back to you.
50:44
And a half hour later they get back and they said, hey, I've
50:46
run your scenario through our system in 100 different ways and
50:52
these are the six that performed best, you know, So we will in
50:56
about a year come out with a platform that would allow
51:00
individuals to actually interface into the capabilities
51:04
of taking voice, doing feature extraction, building it through
51:09
a model based on the design that they have indicated.
51:12
I want to look at this, this, this and this and kind of check
51:15
boxes. And then it'll go through and
51:17
it'll build them a custom little model that they could then
51:22
deploy either on the cloud or actually deploy on their
51:25
computer at home. Let me know, because I'm telling
51:29
you right now, like, I've been married for 20 years and my wife
51:34
has certain tone that in years prior I might walk right into an
51:38
argument. But now I know if I just bring
51:41
her a little bit of chocolate, like, everything's gonna be
51:44
fine. So I could use that information
51:47
to program the robot that if it senses that tone, it's just
51:51
gonna be like, how about a Kit Kat?
51:52
And it'll be like, yeah, let's get it together.
51:54
Sweetheart, Amazon drops it off at the door.
51:59
Alexa automatically just orders the chocolate for you.
52:03
Yeah, no, it's. And and it has been.
52:06
It has 2 profound impacts. 1 is that she gets chocolate and 2nd
52:10
is she knows you cared enough to know when she needed that
52:13
chocolate. That's right.
52:14
That's right. I knew enough to let the robot
52:16
make the decision. We'd just be eating chocolate
52:20
all the time. It'd be great.
52:21
It'd be amazing, yeah. You know, it's nice.
52:25
You mentioned dolphins. I this is probably tangential,
52:30
but like, have you seen the reports that they're like using
52:33
AI to kind of decipher whale calls and dolphin calls and
52:36
stuff like that? Fascinating.
52:38
Yeah, I mean, just like our dogs are capable of understanding,
52:42
you know, 253 hundred, 400 words, whatever it is, you know,
52:47
they're, they're also there's a different bark for fill out my
52:51
water or give me some food or let me out of this house, you
52:54
know? Yeah, there's different barks
52:57
and communications. Well, dolphins, it's estimated
53:00
dolphins are communicating at a level near near ours.
53:04
And if that's the case, AI could be the tool that translates what
53:11
they're saying. I just saw a report where they
53:13
used AI with some cuneiform, you know, Sumerian tablets that have
53:20
never really been well understood.
53:22
And they, they took pictures of these, digitized them and then
53:27
went through and said, what are the patterns that I haven't seen
53:31
yet? Or me as a researcher of 20
53:33
years looking at it, haven't seen.
53:35
And AI is pulling stuff out where we may see the emergence
53:40
of an understanding of that language that we, we would never
53:43
other otherwise have, you know, which I think is really kind of
53:46
cool, you know, I think, you know, go back, you know,
53:49
whatever it was 567000 years and we can begin to understand a
53:54
complex line, complex language that they were writing in.
53:58
Yeah, totally. I know it's so exciting.
54:01
There's there, it's simultaneously there seems to be
54:03
a race. I mean, there's a lot of kind of
54:05
doomer information out there that's kind of scary about AI,
54:08
you know, and that could be right.
54:10
But like thinking optimistically about particularly in medicine,
54:14
like all of the opportunities are that are coming very, very
54:17
soon are are just fascinating. What what inspired you to get in
54:21
the the front line of this? Was it that that bagel meeting?
54:25
Were you like, you know what speech medicine?
54:28
We should do something here. It, it really was.
54:33
Jeff and I were. What kind of bagel was it and
54:36
what kind of cream cheese was it?
54:37
I want to know if there's some I like.
54:39
Cinnamon I like cinnamon Raisin bagels and you can't beat that
54:44
with just butter, honestly. And and so we we were there to
54:49
eat a bagel and we were there ended up being there for 8
54:51
hours. We closed the place down.
54:53
Wow. But I said to Jeff this really
54:56
simple but true story. I said Jeff after he was telling
55:00
me about trying to use words to understand different things,
55:03
where the successes had been and where where there just wasn't
55:06
success, you know, and of course Jeff had created first ever four
55:12
or five different products using voice and speech.
55:15
So I said, Jeff, forgive me because obviously I don't know
55:19
anything. And and we met when he was at
55:23
the NSA and I was at the NIH and I was doing research in
55:27
neurological diseases. He was building mathematical
55:30
models to de encrypt spy messages.
55:32
That's really fun foundationally what speech is, you know.
55:37
And so I said to him, I said, you know, my daughter, my wife
55:41
and I had five kids. You know, my daughter Caitlin,
55:43
when she would come home her junior and senior year from
55:46
soccer, you know, 6 O clock, 6:30 at night.
55:49
And I'm there. I could see when she walked
55:52
across the room by from her gait, whether it was a good day
55:56
or a bad day. You know, when she turned and
55:58
looked at me, you know, I get it.
56:02
And so then I would say, how was your day?
56:05
And I would get the same answer that every parent gets,
56:08
regardless of where they live in the world.
56:11
It was fine. And you keep going.
56:13
You know, when I go time out, you know, come back, you know,
56:16
And so I'm not a hovering parent, but I'm an engaged
56:19
parent. And so she's talking with me
56:21
about her day and we figure it out.
56:23
And I said to Jeff, I don't get that.
56:26
I don't get to do that anymore. She doesn't, she doesn't live
56:29
with us anymore. She's married and she has her
56:31
first child and her and her husband.
56:33
And that's normal. I don't want it.
56:35
I don't like it. I'd prefer she still be here,
56:37
but that's not life, you know? And I said to Jeff, I'll admit I
56:41
call her twice a week now. It's probably more often than
56:44
that, but I'll admit to 2. And I said, Jeff, when I call my
56:48
daughter and I don't have the physical cues anymore and I
56:50
can't see your gait and I can't see your face and I can't see
56:53
your body language. And it doesn't matter what she's
56:58
saying to me. And it's within moments of the
57:00
call starting and it's irritatingly accurate.
57:04
I know if it's a good day or a bad day.
57:06
I know if she's excited or sad or I I know the whole range.
57:12
And I'll say to her, honey, what happened today?
57:15
What's going on? You know, or what are you so
57:18
excited about or whatever it is, you know, and we do that one for
57:25
another. And I said to Jeff, at that
57:27
moment, I said, Jeff, how are we doing that?
57:32
If we could do that in a handheld device like on the
57:37
phone, if we could do that for clinicians and hospitals,
57:41
because they're already doing what I'm doing, but we could
57:44
object. We can make this an objective
57:45
tool. So that's that was kind of the
57:47
aha moment. Jeff's answer was, I don't know.
57:51
I don't know how we're doing. And I said, OK, well, that's
57:53
what we need to figure out. Let's go figure that out.
57:56
So, so that's how it got started.
57:59
Let me let me just tell you one more thing that comes from
58:02
something you mentioned. One of the questions I was asked
58:06
last early or last week on Friday in this conference, I was
58:13
asked the question, The concern is that AI is going to be
58:19
eliminating jobs or that not as many people will be employed or
58:24
it's going to take over for programmers and it's going to
58:27
take over for engineers. And it's going to do this and
58:29
it's going to do that. And they said, what do you think
58:32
is going to happen? What will be the total
58:36
addressable market from an employment standpoint?
58:40
And my answer was that every aspect, everything will
58:47
increase. Everything will AI is going to
58:51
do a whole bunch of things for us.
58:53
But I said, what do you, what do you think?
58:55
I said, do you think when we were digging holes with our
58:58
hands and someone came along, you know, the really bright cave
59:02
person came along with the the first wooden shovel, Do you
59:05
think we dug less holes? Do you think fewer people were
59:08
digging holes? Absolutely not.
59:11
More holes were dug. They built the pyramids after
59:14
that, for heaven's sakes, You know, so and so I said it's AI
59:21
is going, is going to expand the number of people we need.
59:26
It's going to expand the wonderful things they get to do
59:29
and they're going to it's going to increase our capacity to be
59:33
better than we are. I said it's not going to cut
59:37
jobs. It's they're going to be more
59:39
holes dug in the future than there were in the past.
59:43
So that that was my answer. Totally agree, but just briefly
59:46
before we end up here, can I tell you about one of the holes
59:49
I was digging? Sure, absolutely.
59:52
So I was working on a product that was a, it was a voice
59:54
cloning tool largely for legacy for, for our grandparents, so
59:59
that you could capture their voice.
1:00:01
And then if they weren't there, you know, maybe you, they could
1:00:04
read the night before Christmas or something to the grandkids
1:00:07
and, and do fun stuff like that. I thought it'd be like a fun way
1:00:10
to to. Talk to my grandparents
1:00:13
personally. So I was talking to my
1:00:15
mother-in-law about this tool. And I was like, hey, you think,
1:00:17
you know, Mimi would be interested in that?
1:00:19
That was her mother. And she's like, yeah, I think
1:00:21
she really would. She was into it.
1:00:22
And I was like, OK, cool. Well, a few weeks later, Mimi
1:00:25
passed. Rest in peace.
1:00:27
And we're at the wake in San Diego.
1:00:29
And I I go to my mother-in-law and I'm like, you know, I'm so
1:00:32
sorry. I I wanted to do this interview
1:00:34
and I never got to it. I'm so sorry.
1:00:37
I really wish I had. And.
1:00:39
And she said, no, it's OK. I did she's like after we
1:00:42
talked, I wouldn't interviewed her and so we were at the wake
1:00:44
and outside by the pool and she like brought her phone out and
1:00:47
hit play and the characteristics of Mimi's voice that I heard
1:00:52
that I'd never really heard before, like I heard kind of the
1:00:55
frailty a little bit, but also I heard all the humor and what I
1:00:59
really heard was the Philly like you could really hear Philly
1:01:02
thick in her voice that you'd never really noticed before.
1:01:05
So like this kind of like use of voice as a signal, I think has
1:01:12
unbelievable upside and is so much potential.
1:01:15
What it what an amazing company, Henry.
1:01:17
I'm really impressed. You're very generous and I
1:01:20
appreciate that. Thank you very much.
1:01:23
Absolutely. Canary speech.com is the website
1:01:26
where people can check out the product.
1:01:29
What's the best place for people to follow you online?
1:01:33
Well, LinkedIn is a good place online.
1:01:37
This podcast will be a wonderful one for them to follow.
1:01:41
And LinkedIn is the best. I can be reached at
1:01:44
henry@canaryspeech.com. And I answer my emails all the
1:01:49
time. I love it when people write me
1:01:51
and they say, hey, I saw you at this and I was wondering, have
1:01:54
you thought about this? I'd love that kind of stuff.
1:01:57
So later in the evening when I get a moment, I sit down and I'm
1:02:00
answering those emails. I just love it.
1:02:03
I mean, we, I've got people all over the world asking me
1:02:06
questions and some multiple ones, you know, and I asked
1:02:09
them, you know, so it's a, it's you.
1:02:12
You never stop learning, you know, you never stop learning.
1:02:16
And the way you learn best is people ask you something or you
1:02:20
ask them something and you listen.
1:02:22
I mean, the interchange is unbelievably rewarding.
1:02:26
Unbelievable. Well, I got a feeling I'm going
1:02:28
to be emailing you at about midnight being like, all right,
1:02:30
Canary speech is great, but I got some dolphin speech.
1:02:32
Come on, you and me. Let's do it.
1:02:36
It's perfect. Thanks so much.
1:02:39
Appreciate the conversation. Same here.