Share this guide

This is the transcript for the opening keynote presented by Scott Sandland, CEO of Cyrano.ai, presented on day one of Project Voice X.

The transcript below has been modified by the Deepgram team for readability as a blog post, but the original Deepgram ASR-generated transcript was 94% accurate.  Features like diarization, custom vocabulary (keyword boosting), redaction, punctuation, profanity filtering and numeral formatting are all available through Deepgram’s API.  If you want to see if Deepgram is right for your use case, contact us.

[Scott Sandland:] Thank you to Bradley, Rob, the entire Project Voice team. My name is Scott Sandland. I’m the CEO of a company called Cyrano.AI, and what we do is we measure words and language in contextual ways to have subjective understanding about your audience. And just to back up and kinda set the stage, my background is in clinical work. So I had a private practice, and then I was the CEO of a mental health clinic.

In the private practice, I did a lot of high-functioning executive types, a lot of good to great, that kind of work. And then in the clinic that I ran, we worked with a lot of at-risk teams, like drug-addicted adolescents, a lot of people who were in real trouble. And so I went back and forth between those two populations for about twenty years, which is, you know, a considerable amount of time, and I watched the teen crisis unfold in America from a front-row seat. And I’m not gonna make this entire talk about it because I could talk about this for days, but the second leading cause of death in America for people under the age of twenty four is suicide. To the the tune of about three thousand attempts per day just in high schools, and I was watching this get worse, you know, increasing by twenty percent in given years. I mean, just crazy numbers, and then the kids who were getting over medicated or under medicated or dropping out of school, or we’re getting them… to them too late. And I said, we need to do something they can get on the prevention side of this at scale. And so the easy answer is, ok. That’s some form of tech, but what could that be?

So I said, alright. I’m just going to build a chatbot. That’s an easy idea. I’ve trained a staff. I’ll just train a robot, and then that scales better. And so I tried, and I tried really hard for quite a while. And I found out that the systems that were available to me couldn’t do it because those systems were optimized for things that weren’t what I optimize for. See, chat systems, you know, natural language processing systems tend to be optimized for a couple of things.

The first one is efficiency. You know, they want you to get resolved in the contact center fast. It makes sense. Second one is accuracy. Also makes sense. And the third one is complete responses. So complete, accurate responses is what these chat systems are optimized for. And as soon as I really understood that, I went, oh, this is why chat systems are so frustrating because they’re not doing what I do, which is active listening. My peers and my colleagues have a lot of practice with soft skills, with feeling and reading the room, and customizing the message for the intended audience. And everyone in this room knows how to do that, and it’s just about how to break that down into the fundamentals and show the work.

So I went to a friend of mine who’s a sociologist and a neurolinguist, and I said, this is the thing we gotta do. And he’s been in the… that world for about twenty years as well. So between the two of us, we have about forty years experience of subjective assessment of words. So we started building out a system. We did some prototypes that started working well. I’ll show you some examples. We ended up getting a patent in this system that I’m gonna show you today and how it can be meaningful for the people that you guys interact with. Let’s see which button. There we go.

So I love words and communication deeply. I always have.

I believe words and act and effective communication strategies are the thing that defines humanity. In the way that peacocks have feathers and the way that deer have antlers, humans have language, and that language allows us to run this planet. It is not thumbs. It is our idea that we can communicate complex ideas to each other. We can pass on complex ideas to the next generation so we can stand on the shoulders of those who came before us, but also we can support each other. We can help each other. We can empathize. We can crowdsource thought like nothing else we’ve seen in the universe.

The first most valuable thing in the known universe is oceans made out of water. That’s what we’re looking for with all those telescopes and stuff. Oceans made of water. The second most exciting thing is communication. So we’re talking about the thing that defines us is the second most important thing, and we are in a really interesting spot. And and Bradley’s talk half an hour ago is exactly right, that we’re at this really great point in history where the defining technology on the planet is aligned with communication, and that rarely happens. You’re looking at cave drawings. You’re looking at the printing press, the radio, TV, and now as we’re getting into Web three-point-o, communication. And this is only the second time, and this is my hypothesis that as we’re getting into Web three-point-o, what we have is the second time that is not just about broadcasting. It’s not just about how many retweets or how many views on TikTok. It’s second only to the telephone in how we receive that information is just as important. And so I’m genuinely grateful to be on this stage and in this room because some of the people in this room get to be having some of the conversations that make some of the decisions that will define these categories and more for the next couple generations. And that’s a really exciting window for us to just acknowledge.

So if we’re gonna do that, we have to think about how we organize things. We can’t we can’t process all the details. So we have to segment. And and Jeff use a a wonderful example of ’em, but let me back up. First, it started off with demographic segmentation, and that’s a very primal easy thing. People who look like me and live near me should probably think and act like me, and people who look like you and live near you, I will extrapolate probably think and act like you. A very simple and useful thought process, but we all know that’s not good enough, and so we did better than that. And we went to psychographics.

Easy example of psychographics is what Netflix does so well. If you like that, you’re gonna love this because people who enjoy and think about these things, enjoy and think about those things. Also nice. And then you get into behavioral, which is what Amazon does incredibly well. Amazon does, hey. People who buy these things end up needing those things. Or it tells me when I’m out of dog food. It’s… it knows how much my dog eats. It knows the behaviors of my dog and segments the messaging appropriately. Wonderful. But now what if we could add contextual? And the reason I love language is because I believe words are a behavior. I am a behavioralist. I think of words that way as an activity. But the great thing about words is that they are absolutely in the moment contextual because we all say the best indicator of future actions is past actions. But that’s not exactly true because we also have to say to the SCC, past results are not a guarantee of future results. But the present, the right now, that’s accurate. I get to find out how this person’s showing up in front of me today. Not the… their demographics, not if they’re a forty-ish year old white male, but what mood they’re in, what they’re prioritizing.

So we need to move beyond sentiment analysis is the first step of this. No one in this room thinks sentiment analysis is good enough. No one’s looking at sentiment analysis and going, a fifth grade reading level? Yeah. Let’s call it a day. We’ll have billion dollar companies making decisions based on thumbs up, thumbs down assessments. I like the book. I hated the movie. That’s what we got. And there’s some gradients in there, but we need to do more. And so what I’d like to show you is a a small piece of our patent actually is based on this, which is social science established your peer review ideas of motivational interviewing. And we look at the underpinnings of how a person actually makes a decision, and it is this, desire, ability, reason, need, and commitment. I’ll say it again. And, by the way, everything on this presentation is gonna be on the TV at our table back there. So if you wanna see all this, don’t worry about writing anything down or taking good notes. We’ll have it over there slowly, but desire, ability, reason, need, commitment, and then it goes into activation and achievement of a goal.

This is what I would do with my clients. I would pay attention to which one of those were in play and which one of those we needed to augment. So let’s use some simple examples. Dippin’ Dots. Do I want Dippin’ Dots? Yes. High yes. This is not binary. This gets a high score. Do I want Dippin’ Dots? High yes. Can I have Dippin’ Dots? Yes, Bradley, and the good people at Lenovo have made this very easy for me. They’ve made it accessible close and free. So it requires very little agency on my part to have that ability to be very high. Logic, should I have a Dippin’ Dots? It’s kind of a weak score here. If then, if I have Dippin’ Dots, then I will enjoy it, and my blood sugar will go up a little bit. Not really a compelling thing either way. Do I need Dippin’ Dots? No. Do not need them. Am I gonna have Dippin’ Dots? Yeah. Yeah. I absolutely am. You you know your whole heart, I’m gonna have Dippin’ Dots in like an hour. K?

I almost brought it on stage to eat it now. So in this example, high desire, high ability, no reason, and no need high commitment. What’s that come to? Activation. That’s enough. It crosses that threshold, but let’s give you a couple other examples. Because I told you I worked with high-functioning executives and drug-addicted teenagers, and they actually have a similar thought process, which is interesting. High desire, high commitment. That’s it. High desire, high commitment. That is a language pattern that they use. I want to get this done, get it done. High-functioning executive has a team of people who can show their work in the middle, and they trust the people that report to them and their teams to deliver on those middle steps. And so that high functioning, high level, I want us to do this. Let’s get it done. Good enough.

Also, drug addicts talk that way. I wanna get some pills, I go get them. That’s gonna be a great clip out of context. But they have this ability, so I got to watch drug addicts and high-functioning executives have the same language patterns and have the same thought processes. Interestingly, that’s how Donald Trump speaks. It really is. And he’s a high-functioning executive, so it makes sense. But he says we’re gonna do this, we’re gonna make it happen. We will do this. I want this. We will do this. And the high-functioning executives and the people who are hourly employees who are used to being told what to do, both are satisfied by that speech pattern. And this is not a political statement on if it’s a good or a bad thing, it’s just really interesting that the people who wanted and to show his work were unsatisfied by that speech pattern, and the people who were used to that speech pattern resonated with it.

So it wasn’t just the content, it was actually the structure of his message that was congruent with his base, which is why he won an election with a bunch of controversial statements because it was congruent. So the next level of how we can personalize this… and, by the way, what if you had an AI system that knew all that about you and every other person. What if it new desire, ability, reason, need, commitment in real time and changed how it was talking to each person? What if in your HubSpot or in your CRM, you knew where a person’s commitment level was, and you sent them different campaigns based on their word choice? The next way we could analyze a person is personality tests, which is a dramatically flawed science.

This book, Personality Isn’t Permanent in New York Times’ best-selling book, if you haven’t read it, you absolutely should. But the idea is, Jeff, great. You just finished your third interview. We really like you. Your resume is good. Your credentials are good. You got a couple good letters of rec from your previous employer. Now before we give you an offer letter, what I’d like to do is sit you alone in a room, and I’m gonna give you a hundred question multiple-choice test. And you’re going to fill out hypothetical responses to hypothetical global scenarios out of context that we will then apply to you at the identity level for your entire time at the company. And we may even put a thing on the door to your office that labels you as a c or a INTP or an ESTJ, and that’s who you are. But that’s not who you are. You’re not one person. I know that context matters in terms of my priorities. Because if I go to Disneyland with my kid and my wife, I care about relationships, their feelings, experiences, and happy moments. And when I go to the car dealership, I care about bottom-line price out the door. I care about monthly payment. I care about resale value, and my priorities are different.

And I could give you a dozen other examples of that, but I don’t need to because you know how you talk to some people is different than how you talk to others. And how you listen is also different, and you’re paying attention to different things. We’ve all heard that thing. If you want to get money, ask for advice. If you wanna get advice, ask for money. And when you go to a person you say, hey. I need some money. They’re ready to tell you what’s wrong with your idea. Anybody who’s been looking at fundraising with some VCs has had this experience. They put on a different, you know, headspace in a different filter. What if the system was paying attention to your priorities? What if it was segmenting and truly respecting what matters to you? What moves the needle for you? This is what my colleagues and I do all day. This is not a bold idea. That just the idea of how to label the parts becomes interesting. And then learning styles. Quick demo. I’d love for all of you to do this with me. It’s very easy. Everyone just take your right hand and put it straight up like this, please. Now just make it your index finger point at the ceiling. Great. Point at your neighbor. Cool. Now make an ok sign. Everyone take a look at your ok sign. Now look at mine. Go ahead and take and put it right on your chin. k. Real quick. This is your chin. Ok?

So here’s what we noticed. k? This is not an IQ test ’cause a lot of you failed. Right? So here’s what this is. If you did this, you followed my visual suggestion, and you’re maybe a visual learner. If you did this, you heard me say the word chin and did the right thing there. So you’re an auditory learner. And if you did this, you’re a cheater. And I saw you. I saw… you know, you know, you did it, but that’s ok. But we all know there’s visual learners, there’s auditory learners, and there’s kinesthetic sense of feel or touch, and they’re different learning styles. And the way a person processes information, it interacts with the world internally and externally is broken down with this.

Ok. So if you could say, you know, this is a visual person. Make sure the thing is beautiful. Or this is an auditory person, send them a link to a podcast if they… if you want them to learn about this. Or this is a person who cares about feelings, which includes their emotions, so hit them with the heartstrings or give them a sample that they can touch. That’s what matters to each type of person. So this is what our company does. We measure all those things in rank to order and a bunch of other stuff. We do that based on the person’s words and phrases and context so that we know how to change their mind, so we can talk them into and out of decisions, which is a great thing when you’re talking a person into a sale. We’re talking a kid into a treatment plan, or you talk the kid out of self-harm, or you talk a frustrated customer out of churning and going to the competition. You talk to them into and out of these things based on who they are contextually right now, and I should probably show some stats to defend to this.

Without going too far into the details, I just wanna say all of these numbers on this screen right now we’re all done with the same model. The set… we’re not retraining and refactoring for different deployments. One tool gave us all these scores. That top one being in retail automotive where we found a really interesting thing. There were times where a person’s need was so high that nothing else mattered. The desire, ability, reason, need, and commitment chart, they cared about if they had enough need, that was it. We worked with a car dealership ship where there had been a natural disaster, and we were looking at the transcripts. And these people just needed trucks, and so they were calling a Ford dealership. They say, do you have an f one fifty? I need an f one fifty. And they would say, yes. We have one. That’s the ability of the dealership. So we hear about their need and my ability as the dealer. And they go, great. I’ll be down in an hour to pick up a… the f one fifty. They don’t care what color it is. They don’t care at trim level it is. They don’t care what it can… they just… they don’t care. That car dealership could have said, yes. We have an f one fifty, and, by the way, we hate you, and your mom’s ugly. And notice then, fine. That’s great. I’ll be down in an hour to pick up the truck. Their need was so high. But, also, there’s a lot of times where that’s not the case, where you actually have to pay attention to which of these leverage you’re going to pull. And and those numbers in that hundred and twenty day split test was the first deployment of our system. We were hoping for, like, maybe a five percent lift, and the numbers were incredible. So let’s play a game. We all love live demos, so what if we could work at the speed of transcription? Well, then the transcription really matters.

So we partner with Deepgram, who analyzed Bradley about half an hour ago. So what we did… you might know… that’s a picture of him from today. So we really did this. You might have noticed that’s kind of working frantically in the back a few moments ago because we took Bradley speaking, and Deepgram did an amazing job. You can see Claudia right here, making sure everything was working. We took everything he said, ran it through their system. She passed it over to Cyrano. Cyrano ran it through our systems, and this is what Cyrano noticed.

So, again, this will be on the back table, so don’t bother trying to read it right now. But this is genuinely the speech he gave, and anybody who does NLP work knows what this is like. All the colored word represents words within different dimensions within a taxonomy that we pay attention to, and then those words are scaled and weighted in their appropriate dimensions. Ok. So then what do we do with this? Oh, we analyze it. So this is Bradley’s speech. You’ll notice that three and a half minute window where the song was playing is blank, and you’ll notice there’s quotes from those moments and then those two blue points, one where he’s impulsive, maybe a high-functioning executive, maybe drug-addicted adolescent. We’ll see. And then also when he was talking about his family friend and the the little boy, right where he got to resolved, was right at the end of that story.

All of this was created by Cyrano’s analysis, and then it tells about him, his profile. Bradley is a caring and confident individual who cares about relationships and auditory processing. The guy who created a podcast about voice tech is auditory, so shock to us all, but you get some really interesting… he values storytelling, connecting people and ideas, and fairness. Next step is based on this graph, what I need to do next with Bradley. The system gives us bullet point advice so that you don’t need to do any interpretation. It just makes it actionable. Next step, ask them to tailor this for other end users, and they’ll dedicate more energy to the outcome. This tells you exactly what you need to do to be successful. And we know a lot about Bradley based on… we need about one minute of talking is what we need. We can do it with less. We can do it with a couple text messages. By the way, we do this on Twitter. So I’ll show you a couple examples at our table if you want, where we can just take a person’s Twitter feed and create a profile on the person based on what they’re publicly saying ’cause we know a lot about Bradley now. We know how to target our presentations to him. We know how to upsell him. We know how to resolve conflict and deescalate problems, and we know how to negotiate against him without losing rapport. We also know how to show ’em open houses. We also know how to mentor him. We know a lot of other things.

I’m just giving you some high level. But what if this was deployed in more systems and it was able to… and I’ll use an example. We’re working… I’m really excited. I don’t know if Mark’s here yet from Constant Companion. Great organization. They’ve been a speaker a bunch of Project Voice stuff. We’re gonna do some work with them next year that I’m really looking forward to because what if this tool could help grandma. What if when grandma is upset? This tool tells us how to calm her down and deescalate things. What if this helps talk her into taking the medication she doesn’t wanna take or we just give this information to the occupational therapist? And just really simple, hey. Before you go, see grandma. You should know, here are the things to say to help her stay motivated and get her better faster.

What if this system was in your call center so that when you’re still dealing with the IVR chatbot side, this is getting enough data so that if the person breaks containment and goes to a rep, you can route them to a rep who matches well. Or you just give the bullet points to the rep of, hey. This person just need bottom line to get them out of here, or this person is more emotional and needs a moment of connection with you. Make this a connection or don’t. What if that same information could then go to the sales department? And now your remarketing efforts with that customer forever are contextually connected to your product. You’re not talking about their search engine history, you’re not talking about their browsing history, you’re talking about what they care about when they’re using your stuff. That’d be pretty cool.

It’d be pretty exciting if we could say, I know my customer, and I know what matters to them deeply, individually, and personally. We can break people down into a hundred and forty four categories before you start talking about real-time commitment, but no human can track that. When you talk about all the different measurements that we do in real time, my cofounder Dan and I can do that because we had twenty years experience in practice doing it. And we weren’t good at it at first and asking your sales team to remember all the things that I just talked about. And then add, you know, the actual sales pitch and paying attention to the content of the conversation, it’s too much. But the AI can do that, and also the AI can work in multiple languages.

So when you’re having a conversation, you only speak French, and they only speak Chinese, and you’re working through an interpreter, there’s a lot… that’s literally lost in translation. And while you’re both paying attention to the interpreter, whether it be a human or a robot, you missed the soft skills moment, and so this can fill in that layer. So we’ve decided to build this as an API so that we can plug it into literally everything. And we’re gonna be a very narrow, narrow sliver, and we’re just gonna nail this. And there’s a lot of other companies that are looking at the waveform and tonality and pitch rate and all those things. That’s great. Those other companies are looking at pupil dilation and microgesture. Also great. And what we’re looking at is the words, and we’re gonna find the partnerships that allow us to help kids at risk.

We’re gonna find the partnerships that allow us to commercially deploy this, and there’s a lot of logistical mathematical and ethical reasons to start in commercial applications that allow us to do a lot more innovation, and we’re doing that today. And there are some companies here on this list and a few others that I’m really excited to be partnering with because it’s going to allow all of us to get more comfortable with a personal assistant that knows that when I say, hey. Put this in my calendar, that she should put it in pencil. What if Alexa heard me say, yeah. Add that party to my list, and she says, yeah. But, you know, this one could be moved if something else was a bigger priority in their command.

This is where we’re going, and we can get this to the point where we could put these tools on every phone and give every at-risk teenager an entry-level therapist before there’s a problem. At the call center sort of scope of practice, an easy place where we could help a lot of kids. Of course, we could do that with, you know, senior citizens, veterans, and a lot of marginalized populations. And in doing so, create better commercial applications for all of us. So that’s what we’re doing. If you guys want to know how to influence Bradley, you can come to our booth, and you can read that. And we can do this on any of you who like. If you want Cyrano to analyze you and see what it says, happy to do that as well. And either way, thank you guys for your time, and, again, thanks to Bradley for letting me analyze him on the stage.

If you have any feedback about this post, or anything else around Deepgram, we'd love to hear from you. Please let us know in our GitHub discussions .

Unlock language AI at scale with an API call.

Get conversational intelligence with transcription and understanding on the world's best speech AI platform.

Sign Up FreeBook a Demo