SPEAKER_03

If you look at these enterprise voice AI agents, it doesn't understand sarcasm very well. And that's because of how the technology is built.

SPEAKER_00

As far as where we were and where we are today, would you say that that we've made it in voice?

SPEAKER_03

I think we're just getting started. I think there's a lot of problems. Um, but I think voice is just the most intuitive medium of communication. So I definitely see that this is a space that will only get more and more hot.

SPEAKER_00

Voice AI is kind of crowded, but this enterprise grade voice AI that you guys are doing is kind of a different game.

SPEAKER_03

Yeah, I would say so. There's a different way of how we position it. And when we think about truly enterprise grade, we really care about security, that's number one. And then also just when people think about latency, latency is very mission critical for all operations. And you can't have lower latency than what is on your device directly.

SPEAKER_00

So, what makes a voice AI company real versus just maybe something that was well marked?

SPEAKER_03

I think this is where everything is getting very hyped. So I would care less about what are they saying in terms of their solutions page, but what are their customer testimonials looking like? And is that an industry or a use case that's very similar to my own? Then yes, I would be quite interested in that. I think the second thing to also look at is to look at doing POCs and pilots.

SPEAKER_00

Sean Jong is the CTO and technical co-founder of Sonus, an enterprise voice AI company. He is building real-time technology that makes speech clearer, reduces accent bias, and helps humans and AI communicate more naturally across the world. Welcome to Using AI at work. I'm your host, Chris Dag. Each week we'll be learning how today's business owners, entrepreneurs, and ambitious professionals are getting more done with smart use of tomorrow's tech. Let's get started. Right now, every business leader is asking the same question. What are we going to do about AI? If this is you, ChiefAIOfficer.com has the answer. We give you a simple path forward where we provide executive and team training so your people know exactly how to safely use generative AI in their day-to-day. We also manage the deployment and implementation to make sure tools actually get adopted and deliver results. And we'll also guide company-wide transformation so AI becomes part of your operating system, not just another shiny object. The companies that act now will increase productivity, cut costs, and grow faster than their competitors. Those that wait will get left behind. So if you want to make AI work in your business, visit ChiefAIOfficer.com and see how we're helping companies of all sizes finally get results from AI. Hey everybody, Chris Daigle, uh here, your host for Using AI at Work. And um today I'm recording from a hotel room in Myrtle Beach, South Carolina, as we're on a 10-day road tour visiting clients and uh just really helping companies with the adoption of AI in their organization. And one of the things that we don't cover a lot is because we haven't really like uh fully fleshed out the use cases for it is exactly what we're gonna be talking about today with our guest, Sean Zang. Sean is the CTO and technical co-founder of an enterprise voice AI company called SANAS. Did I pronounce that correctly?

SPEAKER_01

Yep, you did.

SPEAKER_00

S-A-N-A-S. That's right. And uh Sean's in the thick of it. He's in Palo Alto. So uh the insights that we're gonna be getting from him today as far as um the capabilities and applications and the future of voice AI are gonna be uh unlike what you're finding on your average YouTube video. So, Sean, welcome to using AI at work.

SPEAKER_03

Awesome. Thanks so much for having me. Super excited for the conversation.

SPEAKER_00

Awesome. So um I guess we'll jump right into it. I I'm uh a strategist around AI, I'm CEO of a AI training and deployment company, but I'm also very fingers on keyboards. I'm very geeky about this stuff, right? And um I think anybody that's uh an a generative AI enthusiast is probably um having fun. And I first got introduced to voice AI by I think it was a company called like air.ai, maybe in 2023, mid-early 2023. And it their demos were great, execution, not so great. But I saw the that the future could be pretty powerful if a company could introduce not synthetic interaction, but what truly felt like authentic interaction via voice over the telephone would be um huge. Agents, voice agents online. I p I I think the the veil is pierced. There's no like suspension of disbelief there, right? But on the phone, um if we had the ability to pull it off, it could be a real game changer for a lot of businesses that can't afford uh the investment of time, of uh resources to have uh people on the phone 24-7 or outbound sales or anything like that. So um you guys have been at this for what, four years?

SPEAKER_03

Uh six years, technically, 2020 was where we got our start.

SPEAKER_00

So um as far as where we were and where we are today, would you say that that we've made it in voice?

SPEAKER_03

I think we're just getting started, to be honest. I think there's a lot of problems. Um, but uh the way that the way the world is investing in it, right? The way that people believe that voice is the new keyboard, I think voice is just the most intuitive medium of communication. So I definitely see that this is a space that will only get more and more hot.

SPEAKER_00

So uh it's interesting you brought this up. I was assuming voice as in like uh voice being the interaction tool between me and somebody else's AI agent or AI model. But you're just suggesting this as voice being the medium through communication with me as a user and the model, not through somebody else's voice, like voice AI, but me directly communicating with it as my interaction mechanism.

SPEAKER_03

I think both, really, right? So I think with speech, you're able to, it's more than just the words that you say. It's how you say it and where you say it and whatever you're using. It's the emotion and your words, the rhythm, and even your speaker identity, right? So when you're speaking with an individual, it's more than just the words. You're trying to portray there's a certain excitement or emotion that I'm also trying to relay into this conversation. And whether that's a human on the other side of that telephone line or if it's an AI agent, I would also hope that that AI agent can understand the urgency in my voice, right? And can understand exactly what is it that I'm trying to achieve, and can I get an engagement that is very responsive, that makes sense and is not getting distracted or off track.

SPEAKER_00

You know, man, this is interesting because I guess I'd never considered that before. I just, I guess I assumed that the models were able to pick up the intonation and the pattern and be able to use that as context for their answer, but you're saying that's not necessarily the case.

SPEAKER_03

I wouldn't say that's the case for these AI voice agents right now. What you often see in their stack is that they're just using the speech to text, right? So they are only transcribing what you're saying, and then you're kind of looking at your words semantically, and if you're saying like dang it or something, then I suppose the AI agent is understanding it just from urgency. But you can actually see if you look at these enterprise voice voice AI agents, it doesn't understand sarcasm very well. And that's because of how the technology is built. And I think these are areas where there can be a lot of improvement because, again, it is more than just the words that I say, it is how I say it, it's where I'm saying it. And there's so many pieces of voice that have so much underlying meaning. And again, that's why voice is so much more powerful than just typing it out and texting it out.

SPEAKER_00

So let me ask have you guys cracked the code on that?

SPEAKER_03

I think we are investing a lot more where we can provide that bigger experience, right?

SPEAKER_00

Yeah.

SPEAKER_03

I'm I'm happy to tell more about Sonics about what we exactly do, but what we try to do is really look at speech disentanglement as our core technology, right? So I keep saying this again, but speech is more than just the words that I say. And what we have built our AI to do is to actually disentangle speech across its different components. So I can isolate, hey, this is your words, these are your pronunciations, and this is your accent, and this is your language, this is your speaker identity, this is your emotion. And what's sudden is what we do is that we try to reconstruct that speech in terms of how you would want to design it, right? So I'll give you an example in terms of background speech or noise. Right now, for if you're calling an AI voice agent on the telephone, it can get distracted and actually respond to my mom on the other side of the living room talking to the TV or talking to Alexa. But if a real human, they would not, they should not do that. They should understand that hey, there's a particular human voice that I'm trying to pay attention to. That voice is also closer to the microphone, not farther away. So, what we do here at Sonus is that we build technologies for humans to understand humans better, but interestingly, also how computers can understand humans better, and more than just the words that you say, but all of those chaotic elements that are inside your speech.

SPEAKER_00

Again, I guess I just assumed that all of the voice agents that I'm working with or models that we're working with or that I encounter are able to do that. But that is not currently a capability that's baked in with these tools.

SPEAKER_03

No, I would say like it can work for maybe the normal, clean speech scenarios. But when it handles those edge cases, I think you will see that it breaks down. Not that these won't get better, right? Yeah. But these are still these nuances where the real world is chaotic, speech is chaotic, right? And I think this is where the companies and technologies need to continue to invest in order for people to feel like that they can start having proper experiences. You know, I often see the landscape to really invest in how do we have more lifelike text-to-speech, right? But to be honest, when I was a kid and I grew up watching all these sci-fi movies, you know, honestly, the robots still sounded robotic.

SPEAKER_01

They sounded like robots, yeah.

SPEAKER_03

Right? But I mean, in those movies, you saw that the interactions were intuitive, right? That robots were responding to you and not getting distracted, were able to understand you accurately. And I think for me, the breaking point with AI adoption is not how natural the voice is, it is how responsive that AI is to your actual conversation.

SPEAKER_00

Interesting. So just technically let me get my head around this. So, in in this environment of progressing capabilities of AI voice, that is a separate model that would still need to interact with a large language model or translate your intention. So this is like a layer that would sit on top of the LLM because the the Sonus model is not going to be the large language model that comes up with the answer and makes the response. Is that correct?

SPEAKER_03

No, not really. Right. So we actually are, you should consider us a bit of like that pre-processing layer, right? Got it. So we're a layer that hopefully can we clean up the audio so that it becomes less distracting with the noises or the background speakers. Yeah. We can also do things like, hey, how can we enhance the articulation and enunciation such that downstream speech-to-text models can understand better? And can speech-to-text models or these AI models can they understand meaning beyond just text? And also, hey, what is the emotion? Hey, what is the background? And use all of that information in terms of coming up with a sponsor. So yeah, we're we're more that pre-processing layer.

SPEAKER_00

Okay, that's very helpful. Thank you. Um so for most companies that you're working with, like tell me who is a who's a without naming names necessarily, but who would be like the typical client avatar for you guys?

SPEAKER_03

Yeah, so I think we so there's both for humans and both for the computers, right? So in terms of humans, we've seen it also help in terms of customer service is a big one, right? Okay. You have customer service representatives, just speaking with customers, and because of all the different environments they're in, they might have trouble understanding each other. So we provide tools to provide clarity in those sort of scenarios. So we do So give me an example.

SPEAKER_00

Would that be because of language or second language or accent, or because of like factory floor background noise or airport background noise? What is the Yeah?

SPEAKER_03

So the beauty of Sonas is that we're a platform, so all of the above, and I'll let me walk you through the different flavors, right? So we have our different accent translation, we have our noise cancellation, speech enhancement, and language translation. In terms of real-world examples that we can use, I think if you're in the office and you have a manager with a thick accent that you personally just can't have a hard time understanding, we want you to think about Sana's accent translation. If you're at a bar with your friends and it's incredibly noisy and it's a bit hard to speak over one another, we want you to think about Sana's noise cancellation. If you're in a wind tunnel with a spotty signal and you hear your voice is being distorted under other lines that telephone, we want you to think about Sana's speech enhancement. And if you're in a foreign country, you're trying to get help or directions, we want you to think about Sonna's language translation. And so this is our suite of solutions. And we all do this in real time with the lowest amount of latency possible, with as much security as possible, trying to do this actually on device rather than sending it to the cloud on the network. And we do this all with the best models that preserve your voice. When you change your language or if you try to translate your accent, we want it to self-sound like you. We don't want it to sound like your Brad Pay or anything because you, who you are, is still extremely relevant to that conversation. So when humans are speaking to humans, you want that emotional connection. Otherwise, yeah, I'm speaking, I would rather speak to an AI agent or rather go through text. So these are these human scenarios that we work with, and that's enterprises, consumers, and contact centers.

SPEAKER_00

So this is interesting because my default association with AI voice was a tool that I used to interact with my clients or engage with prospects on my behalf as a proxy for me having a team, right? But when you're talking about it being on device, are you talking about it being on my phone and I'm actually interacting with Sunas as an app as well as using it to do what I talked about just a moment ago?

SPEAKER_03

Exactly. So we we deploy almost everywhere. So we do deploy as on device on your laptop or your phone. And so you can almost think of us as like a microphone. We're like something that you attach with. And honestly, that's how we position ourselves as. We think of ourselves like an infrastructure layer or microphone layer. Like if you think about, I think noise cancellation is very well accepted nowadays, and people are using it to cancel noise. And you're when you cancel noise, you're not trying to hide the fact that you're not in a noisy environment, more like. And here at Sonas, we just try to take that in the next level and think about okay, there's more than just noise that can add friction points in a conversation. Our job is how do we add clarity to a conversation wherever you are, whoever you are, whatever you're using.

SPEAKER_00

So uh the market, as I mentioned, the first time I encountered this was 2023. Since then, there's certainly been more competitors that have entered the field. Voice AI is kind of crowded. But this enterprise grade voice AI that you guys are doing, it's kind of a different game.

SPEAKER_03

Yeah, I would say so. I there's a different way of how we position it. And I think when we think about truly enterprise grade, we really care about security, that's number one. And data and privacy will only grow more and more important in the future. And so that's why we've been very, very intent. Like we have a lot of purpose in terms of making those models work on device so that your data never actually has to leave your ecosystem and you feel that your data is privately secured. And then also just when people think about latency, latency is very mission critical for all operations. And you can't have lower latency than what is on your device directly. And if we needed to, and there are some users where, you know, I'm already running multiple programs, I just don't have enough space, or I'm using a very low-end computer or hardware. We also provide on-premise deployments as well. So we'll deploy it onto the company servers so that you don't have to feel like it's going through Sonic servers or any other third-party APIs. We want to make sure that your data stays where you are, and that's how we make sure that we position ourselves. And I think that's the future of work. Like local AI is also getting more and more hot.

SPEAKER_00

So I'm going to uh get back to the very specific questions I have in a moment, but I don't want to forget this one. I'm sure that you probably uh are familiar with what Jack Dorsey has done at Block in the past.

SPEAKER_01

Okay.

SPEAKER_00

And for those of you you're listening, you probably heard me mention this over the past several episodes. It's something that we're paying attention to. Dorsey fired, you know, Block let go of 4,000 people in a roughly a day. I thought it was a situation where, oh, AI efficiency got introduced into their business. They just didn't need as many people. Fast forward a month later, he releases a paper called From Hierarchy to Intelligence. Are you familiar with this paper that he's released?

SPEAKER_03

To be honest, not so much, but I would love to learn more.

SPEAKER_00

Yeah, so the idea behind it is that they evaluated the org chart of their company and they realized, hey, a big portion of that middle of the org chart is doing what AI does. They're taking in information, they're evaluating it, synthesizing it, repackaging it, redistributing it up, down, or laterally through our organization. If we could get, if we could replace that layer of the organization with AI, maybe we don't need as many employees. But the issue was it only works if you're able to capture the business artifacts that those individuals were dealing with. And a lot of times it was the water cooler conversation or hey, boss, got a minute, kind of things that that weren't being captured except between those two individuals or the people that sat in that that meeting, right? But Jack said, hey, block is we're a distributed company, all remote, so we're capturing most of that information. So we've got the the context bits that that that intelligence layer would need to be able to replace the analysis and decision making of all those people, right? And so we we've started building this internally, it's become kind of like a a niche uh uh topic on X if you're in the Twitter sphere or the X sphere for AI, right? But one of the things that's gonna be missing that's that's that's an 80-20 of context capture, right? There's still things that that Sonas would catch. Me introducing my back and forth with Chat GPT or whatever my model of choice is, right? Currently, that intelligence layer doesn't have access to that. With this, that would now document uh I guess I could capture both sides of it because as it's coming back from the LLM, is it going back through the Sonus app to verbalize it back to me?

SPEAKER_03

It can, it can, right? So if there ever needed to be an area where you also needed to be able to enhance that clarity going from going outbound or inbound, Sonos can be there to enhance that clarity. And I'll give you just one example is let's say you are speaking with Chat GPT, and again, because ChatGPT is a server-side deployment, you know, Wi-Fi and bandwidth and signal is very important. So if you're in a scenario where you're a very spotty signal, then the ChatGPT voice and your voice might actually break, might be very choppy. You might have these glitches and you start to drop words or phonemes. Sonus, we actually can repair those phonemes, and it starts to sound like a very smooth, continuous speech. And that makes it a lot more engaging and rich of an interaction. So, and and that's also what I think about, honestly. I think about the hundreds of millions of hours of speech that are being transmitted on telephone lines on the internet, and yet there's no real technology being applied in the middle to empower or to enhance or to make those conversations more efficient. And so that's actually where we're very focused on is how can we become like a utility for speech? Like you think about electricity being a utility to uh empower your appliances, we want it such that can saunas be a utility such that whenever there's a conduit of speech, can Sonas sit in and empower that speech to be more clear and be more efficient? Um, so again, everything is just getting started. Speech as a space, I think is just getting started.

SPEAKER_00

Wow, and it's moving so fast. I mean, if you think about what we take for granted now, like I would assume that what you would demonstrate to me would be like, oh my gosh, this is amazing. To you, it's like, nah, we're just getting started. So we're getting started. So here's a question. What what makes an a voice AI company? We've got decision makers of you know, companies of all size listening to the podcast. You know, I'm sure I'm sure if they don't have it active, they're eager to learn more about how could we leverage voice AI instead of the human with all of their foibles and their demands and their biological needs? How can we leverage AI? So let's maybe walk through what they should be doing to make sure that they're not getting a bunch of hype and they're actually getting the right tool for them in this voice AI environment. So, what makes a voice AI company real versus just maybe something that was well marketed?

SPEAKER_03

Makes sense. I think this is where everything is getting very hyped. And so you should not be evaluating a company based off what they're saying. What value what is even more valuable is what are those customers of that company saying, right? So I would care less about what are they saying in terms of their solutions page, but what are their customer testimonials looking like? And is that an industry or a use case that's very similar to my own? Then yes, I would be quite interested in that. I think the second thing to also look at is to look at doing POCs and pilots. Like everything, every new technology, you have to have a bit of that implementation and experimentation stage because you want to be able to validate that this is not just cool tech, but yes, this is a game changer. I see some sort of tangible business metrics that are indeed moving in the right direction. So what I would evaluate is implement the technology, right? But do some controlled testing, do an A-B test and see like, okay, here's my results pre-implementation, these are my results post-implementation, and then ask yourself is that the ROI that I think is very valuable? And I think with AI, you are seeing an AD20. And I think people should be encouraged to explore, like, hey, can I use AI to tackle that initial AD? And maybe that can handle the normal or clean speech or you know, the very simple tasks. But again, humans are complex, speech is complex, and that 20%, I don't think you also want to have a leaky bucket there. And so think about those are areas where those business metrics did not do so good. Then maybe you have to think about all the solutions, all their stop gaps. And so I think is there's no such thing as one solution solving everything, like how everything is a little bit nuanced. I think you as a business owner, you need to also approach your system with that nuance and making sure that everything that you think can cover should be played to those strengths.

SPEAKER_00

Yeah. Yeah, I think that's solid advice for any of the listeners. Like if you can manage your expectations to an 80-20 solution environment, I think you'll be very pleased with AI in general, any tool that you if you're looking for it to well, how come it's not doing the whole thing? Don't wish for the Terminator too soon, right? Like the human still needs to be in the loop. So as long as you as a listener understand that, I think that um you will feel like you're getting a lot of value out of it.

SPEAKER_03

Yeah. And I think even this is advice even for like um AI vendors, right? Is ideally like you should not be the salesperson. You should let your technology do the for you. And and that's what we try to do in Sonnets, is like, yes, I'm not here to sell you, I'm not here to pitch on you. All I ask is, why don't you deploy my technology and let the metrics and numbers do the selling for you? That's how we train our team. And honestly, that's how all technology should be evaluated, and especially because the space is getting more crowded. Everyone can be a great salesman, everyone can have a great ad saying they're the best. The proof is always in a pudding. You have to make sure that your product is the one that can actually shine.

SPEAKER_00

You know, uh, and and I'll concur with that just from our side of the thing where we're offering, you know, like services. Um the people that we're talking to, they don't need to be sold on AI. They need to trust that we're a resource that can get the job done, right? Because there is, like you said, there's there's a lot of noise. And so this is good. Um, and so for those of you listening, like tell them you don't need to see this the the product page, you want to see the demonstration as Sean just suggested. I think that's fantastic, and we're gonna actually introduce that more into our own sales process. So good idea. Um so okay, on the on the voice side of things, what are maybe some technical or or business benchmarks that uh decision makers should be asking about before they would buy any type of voice AI solution? Makes sense.

SPEAKER_03

I think it's still always gonna fall into the same fundamental business metrics, right? So if you're thinking about customer service or these call interactions, the same ones would apply. One, customer satisfaction. Are people happy actually taking these calls? Well, that's a human or it's an AI agent, right? Are they satisfied? Two, it's gonna be on call conclusions, right? Are calls actually being concluded, they're not being hung up out of frustration. Are things actually tasks actually being completed? And then three is going to be average handle time, right? If the average handle time is shorter and calls are being completed successfully, then I assume customers are being more happy because they're being solved faster. Your volume is able to increase. And so, regardless of what technology, like these are the same core pillars and metrics that your business revolves around of, and that's what you need to evaluate on. It's not going to be the fashion flashy buttons or the flashy UI UX. It's going to be what actually makes the dollars continue to grow and matter.

SPEAKER_00

Yeah. Yeah. So in in the spirit of this 80-20 conversation, where should somebody who's evaluating a voice AI solution, where should they expect that these AI products are still breaking down in practice?

SPEAKER_03

Makes sense. So I think from what I so I think there's one thing of what I see today, and in there two is also what I think in terms of my thesis for the future. So I think what you see today is that you will see a breakdown in the non-clean scenarios. So there are people with particularly foreign accents, and you know, AI has its own speech bias problem. You can see like Alexa never works for my mom. So there definitely are accent biases that are there. Noise biases are definitely there where if you're in a very noisy scenario, it's starting to be inaccurate in getting your words, or maybe it's fixating on not you, but other speakers around you, right? And then three, there can be other problems in terms of reliability, whether that's the Wi-Fi signal, or maybe it's the actual server service of the vendor themselves, right? And so I can see, and then also the other thing is it's going to be the actual task at hand. Sometimes things require more nuanced scenarios. Yeah. Not all customer is the same, not all problems or complaints are the same. And those edge cases start to become very important. And you also don't want your AI to start making approvals that you would not approve and start having things go chaotic. So maybe one analogy that I'll say is I often liken voice AI, whether it's humans or computers, similar to ATM machines. And what I mean by that is when the ATM machine was invented, a lot of folks maybe were like, hey, we don't need bank tellers anymore. Everything will be automated instead. Yes. Yeah, yeah. What you see today is that there are more bank tellers now than pre-ATM machine invention because one, the market grew, there are more banks around, but then also that bank teller role also evolved. It became less transactional, more of a face of the company, more of a way to build a relationship with a customer. Or three, is like, how do I handle a complex business decision for me to make a human supervised uh approval? Whether that's a complex bank loan or something else of the like, right? So I think that's the interaction that you want to have. I don't think you want people to handle it, if it's a very simple case, maybe it's like taking a drive-thru order. Maybe you're hoping that the AI can cover it. But if not, maybe the AI can tell you that, hey, I don't know if I'm accurate. I want another human to be able to supervise me. And then you want a human supervisor to maybe look at and approve or handle that request. So no matter what, customer service will always be king. And just because you can handle 80% of a customer, but that final 20%, even if it works for that customer today, if it does not work for that customer tomorrow, your brand of quality deteriorates. And so that reliability, it becomes more true than ever.

SPEAKER_00

Sean, are you familiar with um Klarna and what they did with OpenAI early, like maybe 2023, where they they brought AI in to support their uh client services, right? Um as I understand it, smashing success fired most of the team. The net promoter score from the from the customer side was at parity, if not, you know, favoring the AI interaction. Everything looked positive. And then fast forward, they ended up bringing back some or most of that staff. Is that how you understand what happened?

SPEAKER_03

Yeah, I would say so. I think they probably had yeah.

SPEAKER_00

So I was gonna ask me ask, like, what's your take on it?

SPEAKER_03

I think this is where when they did certain POCs and you had the metrics, it looked probably very strong on some subsets, and you're like, hey, we see maybe it's 80% of success, but we're able to reduce our cost or efficiency so much lower. But then again, with that 20% was churning, like right now it's 80% of your customer base today, and then 80% of that 80% is 64%, right? And then you start dwindling, and then what happens is that you become a brand that you're reliable today, but I don't believe that you're reliable tomorrow. I'm gonna look at all their solutions in the market. And I think this is where when we had that previous topic, when you do your POC, evaluate across all the dimensions and don't just look at one metric or one pillar of your business. You have to look at everything, and I don't believe that there will be one solution for all. No technology has ever had that. All technologies have some limitations, but it's up to you as a business owner to be like, okay, I have multiple options, I have multiple aspects of my business. What do I need to do to orchestrate it together to be able to get the best experience for my user and customer base?

SPEAKER_00

Okay, now I know one of the things that you guys are particularly strong at is the gap between my query and AI's response, the delay, the latency, right? And I know that like in in natural language, uh standard uh reply or gap between the reply is what 650 milliseconds roughly.

SPEAKER_02

Around there, yeah.

SPEAKER_00

Something like that. And now, yeah, but you guys have actually achieved a faster response time than 650 milliseconds, is that right?

SPEAKER_03

Yeah, about 200 milliseconds for us.

SPEAKER_00

Why so for the listeners, I I have a good idea, but why does that matter?

SPEAKER_03

I think it matters just because you want to have engaging responses and interactions. Well, again, whether it's a human or computer, so I don't want to necessarily have that awkward silence because that just feels unnatural. And again, no matter how realistic that text-of-speech voice is, I know it's I know it's an AI. And it's okay for it to be an AI, but I want my interaction to feel natural. I want it to feel organic for myself because I'm an organic person. I myself as I'm a human. I enjoy having my experiences to continue to be human and intuitive in that sense. So that's why laten no one's ever gonna say, can I have more latency, right? That's why no one's ever gonna say, like, can your audio be worse? Can you sound more noisy? No one's ever gonna say that. These are gonna be these are theses that will withstand all generations. And so that's where we continue to make investments. That's how we can need to continue to make improvements.

SPEAKER_00

So a big part of that that favorable latency on your part has to do with the fact that it's on device and it's not going somewhere to process and then returning? Yeah, that's a big part of it. That's right. Yeah. Now we we've talked about customer service, and I totally get that, especially if it's phone heavy. I know the impact that that can have. That voice AI can have. A couple of questions. Do the clients that you're working with are they obligated or encouraged to say, by the way, this is an AI agent that you're dealing with?

SPEAKER_03

Yeah. I think from what we see, we kind of leave it to our customers and potentially the regulators, right? So it's up to them whether or not they want to say this is regulated, or I mean this is being used, these different AI tools, or if it's not, I would say it's up to them. And the way I see it is we're here to provide more of that freedom and liberty. I think more freedom is more power. And that's how we want to also express it to users. In fact, when we deploy our technology across different users, we're even like we recommend giving them the choice. Some people might be skeptics to adopting AI. And maybe you have some groups that use AI and some groups that don't. And for those that maybe don't, but they see positive results in the other group, they might build FOMO. And it actually helps encourage adoption or at least still have a discussion and they'll teach and learn from one another. And I also think for Sonsa's case, we kind of see ourselves a bit more like again, like that microphone. We're not sure, would you tell folks about what kind of microphone you're using or whether they're not using noise cancellation? It's it's it's up to I think the vendor, right? I think what we're focused on is how do we improve the clarity, and we are willing to partner with everyone to make that happen.

SPEAKER_00

Nice. Um, do you have any uh clients or or use cases or case studies where um using it in a sales, particularly an outbound sales environment, has been successful?

SPEAKER_03

Yeah, we have definitely.

SPEAKER_00

Okay. To me, that's like the customer service is obvious, right? But the the outbound sales, because uh I don't know how involved with a sales team you are, Sean, but sales sales professionals can be princesses sometimes. Um getting them to follow and report and all that type of stuff isn't like I just generally that seems to be the experience with with sales teams, right? And boy, if I could have a sales team that could do as well as them, but it was AI generated, I might like to try that. So you're saying we're there.

SPEAKER_03

I think we're there in terms of different ways you want to support it, right? So the way we what we do here at Sonics is that we're here to again provide clarity to those conversations. Now, I think clarity is very important because clarity is also a representation of your brand. If I am reaching out to you and I'm trying to sell you my product, I don't want you to get distracted by like, hey, this guy is calling me from where? Like this guy is, I mean, it's very noisy on his hand, or I can't understand this, right? Or I don't even speak the language. I don't even understand, I can't even fundamentally understand what's going on. So it's all these different elements that I don't want to be able to have my customer being distracted. I want to be able to show them, like, hey, this is why I'm reaching out for you for this call. This is what I want to be able to portray to you. And I think clarity is how you're supposed to enhance that communication for both your sales, your seller, as well as the buyer, right? Because I also don't want my sales team to also start getting fatigued working eight hours a day, and you also don't know where those different colleagues are going to be. And it can be very also difficult for you to understand them. I don't want my sales team to also be like, hey, uh, what I didn't catch that. You know, can you repeat that? Because at that point, the colleague's gonna be like, I don't think this person is engaged with me. So no matter what, you want to provide that organic, that intuitive, the engaging experience for both sides of the party. Yeah.

SPEAKER_00

This this concept of I I want the latency, but I want the accuracy. Should um somebody who's getting voice AI should they expect that there is going to be a drop? I mean, because depending on customer service, maybe it's you know, there's an acceptable tolerance of misunderstanding or whatever. But in certain industries, I would imagine, if they wanted to get is it accurate enough to where certain industries that need this type of like uh zero tolerance for error can use this stuff?

SPEAKER_03

I've seen uh I've seen some industries where it'd probably be too early, I would say, right? So there might be some areas where you still want that human in the loop because you are dealing with very critical experiences. Like I'll give you an example. Let's say you're a doctor and you're prescribing something. Yes, the difference between 100 milligrams and 1,000 milligrams, I mean, that can be life or death, right? And so are we truly at the dare for us to rely on AI? I would be nervous about it. And again, the problem with AI is that it's sort of very early, and if it's only transcribing speech to text, it may not understand the urgency, might not understand sarcasm, and if it mistranscribes it because of noise or something, that becomes very dangerous, right? So those are areas where again, you just want to evaluate the business and the you know very holistically. And I think the other thing is there might be some areas where, although it can maybe be less accurate, it's something that it's so urgent that it might be necessary. And maybe I can think about in terms of language translation. Like if I was doing a 911 call, but I did not I did not speak English, but my operator only spoke English, I want to make sure that I can portray my emergency as best as possible. And you know what? Maybe AI is the best way to be able to do that, and I'm willing to take that risk, or there's more metrics from that AI that's able to say, like, I'm 80% accurate, I'm 80% confident, or just all the metrics involved. Like, I'm hoping that AI can also be an enablement in that regard as well. So just nuances, I would say.

SPEAKER_00

Yeah. Yeah. Fascinating topic, and I'm glad to know that that we haven't solved it yet. Right. I'm I mean, congratulations on being ahead of the curve, but I'm glad to know that there's still more progress to be made with voice AI. So, what's next for Sonas? What are you like, what does the future look like for you guys?

SPEAKER_03

Yeah, definitely. So I think on our side, we are also looking to spread our wings where we got our start in the customer service industry and we love the industry and we continue to do very well, and we will continue to go deeper with our partners there. But our business has always been how do we become how do we a communication technology company, right? Our job is how do we make those conversations wherever they are sound more clear so that humans and computers can understand one another. So on our side, we are branching out and also thinking about hey, how can Sonus sit more in the telephone network directly? How can it sit in terms of your phone or your headset? We've also been expanding in terms of enterprise communications, more in terms of even externalizing our models as SDK so that all the enterprises can use our own capabilities wherever they want to do so. So on our side, is we're we're very excited and we continue to love the CX industry, but I mean, communication is everywhere, right? And I think about there's 8 billion people in the world, population size will only go up. I think globalization continues to increase. There's going to be more international work and more communication happening digitally. I think these are all outlets where voice traffic will only increase. Yeah. And because of that, I think there's even more service area for Sonics to embed ourselves in. And we're very excited to do that.

SPEAKER_00

It's a much more expansive future than I was thinking about for sure. I'm sitting over here thinking about it simply as a mechanism for interacting or even not deceiving, but but you know, uh uh sounding authentic enough that a human can interact with it, but you're talking about something that becomes a key technology for global communication, which I like the sound of that. Rashad, thank you so much. So um I like if somebody's on the phone right now, like who would you want to talk to if they were gonna reach out?

SPEAKER_03

I mean, I would love to talk to everyone. I mean, I would love to talk to I think people in terms of their business and they're still thinking about okay, voice continues to be my main inter interface of reaching out to customers or users, I would love to talk to that individual. If you're a company that you do a lot of enterprise uh international communication, I want you to think I want to be able to reach out to you. If you're a cust if you're a business that you always think about quality and that clarity is very closely tied with that quality, we want you to think about Sonas and we want to be able to work with you.

SPEAKER_00

Very cool, man. Thank you for um taking time out of uh the high speed activity of Silicon Valley. But uh we appreciate it, Sean. And thank you so much for sharing your insider's perspective on um voice AI. It's a topic that if if executed well by you know the company that's using these tools, like really opens up a ton of bandwidth in certain areas. So I'm very bullish on it in the future. We're not using a ton of it in our company yet, but after this conversation today, I'm encouraged to uh get more serious about it. So thanks again, man.

SPEAKER_03

Awesome. Thank you so much. It was really fun for the conversation. Thank you.

SPEAKER_00

And everybody listening, thank you so much for joining us. Uh, we'll have another episode out next week. If you got some value out of this, you've got some uh peers or friends that are on the AI journey, and you think that the way that we talk about AI, the things that we talk about could be helpful for them and just getting really like a holistic understanding of what this new paradigm of business looks like. We'd love for you to forward the um the podcast to them and we'll welcome them as guests. Uh tune in next week for another exciting episode of Using AI at Work. Thanks, everybody. Thanks for tuning in to Using AI at Work. Don't forget to subscribe for more conversations about how to use AI at work. And a special thank you to our sponsor, Chief AI Officer, for empowering businesses with AI education and training. Visit their website for a free AI readiness assessment and AI strategy guide to help you get started using AI at work. That's www.chiefaiofficer.com. Follow us on Twitter at the handle UsingAI at Work and visit www.usingai at work.com for free resources to help you harness AI in your role.