Building a Speech AI Platform for Human Communication

Hey everyone. Welcome back here to Techstrong TV. I'm really happy to introduce you to my next guest.

It's his first time here on Techstrong TV, so let's give him a hello. His name is Shawn Zhang, and Shawn is the co-founder CTO of a company called Sanas. Shawn, I hope I got that all right.

No, you got it perfect. Yeah. Thanks so much for having me.

Good, man. My pleasure. Welcome here to Techstrong TV.

So, before we jump in to Sanas and everything, let's hear a little bit about Shawn. Yeah. I know you're out in the valley today in Palo Alto.

How'd you wind up there? Yeah. So I would say I've not traveled very far since I was a college student.

So I entered Stanford University in 2018, spent there around two years. I was studying in computer science, also specializing in AI, was also doing work as a research assistant in Stanford's AI labs. And then in 2020, that was when we had started Sanas and a little over six years from then, here I am.

So, very happy to continue to do what I love to do. So let me ask you a question. You actually started it out of the lab, or the work you did in the labs led to- So I would say- ...

to this company ... the lab got me the foundational discipline, I would say, but truly, us starting Sanas was more of a personal project. And the reason why we started Sanas in the first place was actually inspired from my friend, my classmate, and his personal story.

So I have a friend, and he was also a Stanford student, but he had to take a leave of absence. So he had to be physically present with his family in Nicaragua, and he took a leave of absence. Hmm.

He moved there. He also wanted to help financially support his family, and so he got a job as a local contact center agent. And I was just catching up with him, "Hey, man, how's life?

" And he told me how much he was hating that job. There were a few elements that I think were not too surprising, but for me, what I was super surprised by was he was most frustrated about complaints on his accent. Right?

Mm-hmm. " These would escalate into arguments, and he was getting basically berated and abused every day going to work, speaking on the phone 10 hours a day. Oof.

Right? And as college students, I think that's how we got exposed to the contact center industry, and that accents were a huge friction point for them. There would even be recruiters that would filter applicants based off how thick the accent was, and obviously we thought that was incredibly unfair.

People should be evaluated on their work ethic, their talent, or intellect. They shouldn't need to change their accent for them to succeed in a job that they wish to do. So it was because- As someone who doesn't have any accent- Mm-hmm ...

right, I empathize. But seriously, I'm reminded of, I went to law school- Uh-huh ... 100 years ago, and I learned that my professor in civil trial advocacy was this guy, Harry Lipsig.

Mm-hmm. He was like the dean of tort lawyers, Lipsig and Sullivan. He literally wrote the textbook that you use for torts, one of the textbooks.

And Shawn, he used to take all of his young attorneys- ... and send them prediction lessons so that they would speak with a British accent. Because people on juries, primarily women, and I don't mean to be sexist or anything- ...

but primarily women. Women find British accents- Yeah ... appealing.

Yeah. Appealing, right? Ask Sean Connery.

But they find it appealing, and they tend to believe- Mm-hmm ... whatever they're saying. It's very believable.

Yeah. And when you're a lawyer, that's a good quality you want, right? You want people to believe what you're saying.

Definitely. So, I know firsthand that accents absolutely have a huge role in the impression people get and what it does. So this is really a good thing that you're doing.

But you guys are at it six years. I'm sure over that time- Yeah ... things have changed, grown.

That's right. So I think, so accent is where we got our start. It was our inspiration, it was our- Mm-hmm ...

origin story, and it was a lot of that area about how do we remove friction that can get in the way of people and people understanding each other. But I will say with Sanas, we also want to spread our wings. We also want to become more than just an accent AI company, but an overall speech AI company, even more than just the original contact center.

We didn't want to be just contact center software, but overall a speech platform. So here's our mission. Our mission here at Sanas is how do we help the world understand and be understood?

Because there's a lot of elements in communication. Beyond accents, there's noise, acoustic environments, there's language, there's linguistics, there's network connectivity issues. There's a lot of friction points that can get in the way of communication.

So what we want to use with our AI is how does everyone, everywhere, whatever they're using- Can always be very confident to feel understood and to understand each other. Right. And so we've also expanded our suite of solutions, going beyond just accent translation.

We also have our speech enhancement, language translation, speech intelligence, and it's all of a platform that all revolves around speech. So that's where Sonus has evolved and that's where we're continuing to grow. I love it.

I love it. And it's so funny, right before I got on here, my ChatGPT- Mm-hmm ... upgraded, updated to the latest version.

And so this, I guess it's ChatGPT Voice. Yeah. Yep.

I don't know if you're familiar. It came on, and I could pick what voice and everything, what accent the voice has. And I think, for as long as I've been in computers, the human machine interface has been, we've talked about moving on- Yeah ...

from mouse and windows and keyboards to maybe some sort of voice activating, Star Trek, right? Yep. And I think really now with AI, we're probably closer than ever, let's say.

I'm not saying it's going to happen tomorrow, but we're closer than ever to changing- Mm-hmm ... the machine-human interface to a voice interface. Yeah.

And it'll be interesting to see if our own biases make their way into the AI or whatever that's working these voice interfaces- Mm-hmm ... and this kind of pattern continues. But I'm just, because I'm a Star Trek fan, I'm throwing it out there.

You're an expert, Shawn. What do you think? What does speech AI and the future look like?

Yeah. I think first thing I'll say is that I'm really passionate about speech because I truly see it as the most emotional medium of communication, also the most collaborative medium of communication, right? Mm-hmm.

It's just very natural for us. I think it's very funny that we, as people, have evolved to now be in front of computers, be typing, when technically that's a little bit unnatural to biology, right? But meanwhile, I'm speaking to you.

Our listeners are listening in. I'm speaking to my friends. I'm not here to be typing on a chat board, right?

Texting and everything. I'm here to talk because it's more than just the words that I say, it's also the way I'm saying it. It's also my speaker identity that my mom could recognize that the voice is mine.

So I'm really excited because I think with AI, we are able to unlock and tap in back to our roots of how humans naturally communicate. And then I see two things. I think one with AI, there's a lot of focus on automation, right?

I think here at Sonus, we like to approach things in a different angle, which is how do we really have human-centered AI, right? So first thing's talking about our existing suite, which is accent, speech enhancement, language translation. These are all focused on human-to-human interfaces, like how can humans speak to one another using AI to assist in the communication so that both people can understand each other clearer.

But even with AI, I think there's a big push on automation, on human-to-machine interfaces. The way I see that is let AI be a tool. Let AI be something that allows me to complete tasks easier and faster, so that I have more time for human-to-human communication.

An analogy I like to use is ATM machines in a bank. I'll want to use an ATM machine if I just want to get a task that's very clear, very quick, done. But if there's something that's more complex and I want the human ear, this is a complex, this is a high-empathy conversation, I think I want that human touch, and I want to go into the bank and have a conversation, right?

So I think that's how we want to build AI, not just for the sake of technology, but how does it improve our human interactions? And how does speech AI get us to that future? Let's dive in, Shawn, though.

I wanted to dive into a couple of questions around this whole issue of AI and speech and everything. Mm-hmm. AI will help humanize agents or help human agents, not replace them with- Mm-hmm ...

AI agents. Yeah. Prove me wrong.

Right? I think a lot of people would have a problem with that- Yeah ... statement.

I think how I see things is, I'm just going to piggyback off of that ATM analogy, right? When I think about- Go ahead ... customer service and those conversations, my thesis is that L1 and L2 conversations, which are low complexity, low empathy, yeah, let them be automated because I don't think you want to be on a huge wait list for you to just do something very quick.

Like you just want to schedule an appointment or something, something that's very black and white, whether or not you're getting an outcome or not. But for L3 and L4, which are high complexity and high empathy conversations, I think you do want that human touch. And the example again is ATM machine, L1, L2, very clear-cut of what I want to do.

But sometimes, let's say I need to ask for a mortgage. I need to get a specific bank loan. In that, I need to make a particular appeal because someone and some human should be able to make an approval.

Someone should make a decision and hear my case out, and ultimately make that decision whether or not to approve my particular case. And for something like that, I'm not here to testify and appeal to a machine. I'm in a very human scenario.

It's very personal for me. I want to make sure that I feel heard and that some other human would be able to empathize with me as well. So that's how I kind of see it, and that's the future.

Even with my conversations with enterprises, I think there was previously a huge push about AI is going to automate your contact centers, it's going to automate your human operations. But when enterprises actually adopted, they were not happy because what happened is that maybe I can make 80% of your calls happy, I'll make 80% of your customers happy. Then I'll make 80% of your 80% happy, and then you start having a leaky bucket.

So I think let AI see where it can automate and help quicken some things, but don't lose the human touch because that's the brand you want to be able to build as a company. And I think you want to build an operations that can manage both the human and the technology-assisted areas very harmoniously. Fair enough.

One other issue. Deep fakes, security. Security's top of everyone's mind these days.

What can we do about that? Yeah. I think definitely there's ways to build responsible AI.

So I think there's one, there's ways for us to be active on companies to make sure that they have an obligation there. Two is that I also do think we need to build technology to defend on this as well. So when I think about telephone carriers or other companies, I think it will become almost a standard to build tools to try to automatically detect, hey, this is a deep fake voice, this is a synthetic voice, this does not sound like authentic voice.

Let me at least give a warning to the user that's calling it. Right? And the reason why I'm going to pitch that that's going to be a future that AI has been awesome, but it is increasing more attack vectors, and we need to be able to cover our surfaces.

So the way I see it is, if I think about telecall, and the telephone carrier network that you might be using, I think I would persuade such companies that you should build in defenses to determine whether or not this voice seems authentic or not. Very much how an email we have spam, right? You can have warnings of alerts that, hey, this might be a phishing attempt.

I think we need to put that in the voice channel. And we, as a technology, because we're going from text-based traffic to voice-based traffic, we are also obligated to be responsible to protect not just ourselves, but our users, our customers, and everyone there. " Your mother, your grandmother, we're going to be able to provide guardrails to make sure that they feel protected and that they won't become vulnerable to the rise of AI-driven attacks.

Fair enough. Shawn, we're almost out of time. Just real quick, for people wanting to get more information on Sanas, where do they go?

ai. We have our website there. And go to our website, reach out to us.

We're always happy to be able to listen and learn more and see where can we provide more value to you, and where can speech AI come in, improve your operations, drive more impact, and just have you communicate a lot easier. Love it. Hey, man, thanks for coming up here on Techstrong TV.

Continued success. Come back and keep us posted. Awesome.

Thank you so much. Thanks for the opportunity. My pleasure.

Shawn Zhang, co-founder, CTO at Sanas here on Techstrong TV. Thank you, everyone. We're going to take a break.

We'll be right back.