Hi, my name is Doctor Patricia Scamman. I am the founder and executive chair of Soapbox Labs. Um So at Soapbox, we build speech recognition technology for Children.Um We have, we license our third party software to third parties to integrate into their products and um services and they bring those products to market. So we are basically like the Intel inside for um third party products and services for Children. It was very core to what we built was that we were building for uh Children. Um the demographic that is, it's particularly sensitive and when we talk about A I and data, we have to be particularly sensitive particularly when it comes to our Children. Um So my own background, I won't go on too long with my own background. Um I'm an engineer. I have a phd in speech recognition. Um and A I and I've been in this space for about 25 years. I've worked across academia and industry before founding Soapbox Labs in 2013. So I'm not uh able to share the slides here for some reason. I cannot, it just will not work. I'm gonna try one more time. Um And if I can't, I'm just gonna keep talking and I think that's probably the easiest thing to do. Uh Let me see, can I do this? Let me try this, see if you guys can see that. Yes, I think you can see that. Ok, great.

So I'll keep going. Um So what I want to talk today is about A I for kids and unlocking the potential for A I to ethical use. And of course, it's really important when it comes to kids. So voice technology is the future of human computer interactions. We've always talked about human computer interactions. We've used mouse touch, swipe uh tap in order to interact with computers. But ideally, we'd like to use our voices because that's how we communicate with each other, how humans communicate with each other. So what we're seeing with voice technology and Alexa and Siri in the home is just the tip of the iceberg. We can imagine in the future, um how voice technology will just become the default mode of interaction with technology and now that it's working to a point that it is useful. So we'll see it across education, utility media, gaming, and particularly in the metaverse. So just to tell you a little bit more about soapbox, we found it in 2013. Um The view was to build speech technology that actually works for Children. It was my own experience with my daughter when she was not even four years old. Interacting with educational apps and services that were trying to teach her to read, but really missing that component around um voice about able to listen to her reading.

So while they were doing a great job of teaching certain aspects of reading, what really struck me was that it was missing the element of allowing the child to read, to vocalize, to do that expressive voice and giving her feedback on that. And if you think about it in any language learning, any kind of reading project was really essential that led me to thinking about, wow, how much more essential is speech recognition for Children even more than adults? Because there's more of a need there when they're preliterate and they don't have the complex thinking of menu systems. So we set out to build the highest performance speech recognition for Children. Um We have a world class team around A I and machine learning with 35 people here in Dublin. And what we've had to do is reimagine every step of the way in creating a speech recognition for Children. So think about how we were doing it. And for most of my career, I was building for adults and how we would do this for Children. Um We've been independently validated for a lack of bias and privacy by design and I'll talk a little bit more about that. So first of all, why did we have to build speech recognition for, for Children specifically but like I said, when I was watching my own daughter and interacting, using her voice or trying to, at the very least, I'd spent my career working in adult speech technician.

And I was able to see very clearly the differences between adults and Children physically in terms of their vocal track length and thickness, vocal full size. And you can see here on the right um that actually this is what how the word umbrella sounds like in the frequency domain. And even when you take out any factor of uh volume, how different it is for a four year old, a seven year old and a 10 year old, an adult. And you can see then and imagine how a system that has been built and trained and designed on adult voices and adult language and adult behaviors starts to really struggle when they encounter different physical voices, different languages because language rules because kids don't follow language rules, how erratic and unpredictable.

And if you think about how kids speak, they don't tend to speak like adults. So what we found is the younger the child, the greater differences, which meant that systems that have been built for adults trained and modeled on adult voices and behaviors and language start to break down and not work well for Children. So what's really important here is that when we have modern speech recognition systems that it's working accurately and that means removing the false negatives and false positives. So not telling a child they're right when they're wrong or they're wrong when they're right. Um, is really key.

So if you're telling a kid, they're right when they're wrong, it's not educational and it's going, they're going to know it and when they're wrong, when they're actually right, it's really going to damage confidence. And you can imagine that in so many different scenarios and that's what we mean. When we say high accuracy for Children, we've been independently validated by, through many of our clients and one of which is imagine learning to show that our system works as accurately as human assessors and for Children, but also for diversity of accents, dialects and learning abilities.

And that's really, really important when you're bringing a product to work in the market. Um What we've seen in the market also is that a study that was released and covered by the New York Times last year showed that there's a racial divide in that Alexa Siri, uh Google were all performing at times about 35% lower accuracy for black speakers than white speakers.

And that's huge. If you think about bringing technology, that's bias, that's giving more errors like false positives or false negatives to um certain kids in the in the classroom. I mean, that's actually creating even worse problems. So what you're trying to do is bring A I into the classroom in order to solve problems to be able to bring equity into the classroom. So you've got to be really sure that those systems that are brought in are not creating the worst problem. So our system has been shown to have lack of bias for different courts and different demographics, which is really important. Another really important thing around ethical A I is data. How do you deal that we built that in from the beginning, data is never shared with third parties.

Only used to improve the speech models never used for marketing or product placements. And we set that bar and that red line is there for us and really, really important to build trust in the system. Any data center system is anonymized and deidentified. And we've been independently validated by Privo, which is a safe harbor in the US since our inception, since our founding in 2013 to adhere to GDPR and C A compliance across and other compliance across the globe. So just a quick example because I'm we're short on time now is just to look at the, the voice A I use cases for kids. So on the right, we can see it is being used today in science, maths, English language learning, reading, fluency assessments, keyword spotting. And on the left, we can see it's been using gaming, wake words, transcription, voice activity detection, commands and conversation as well as accessibility. So you can see all these use cases that really bring such utility to children's lives, whether it's in play or whether it's in learning the value um unique value of voice data for kids that can give immediate feedback, they can say something that something happens, they can read something and get feedback for teachers and schools.

They can know how kids are progressing which kids need intervention. And when for researchers, we can see what's working and what's not working in terms of different pedagogy approaches. And for education and media companies, we get feedback on, on progress and where funding is needed and where intervention stepping in is needed. And all these things can really augment the experience for Children, education and product progress in the field. So I'll leave you with this quote um that we, we've really, it's very hard to imagine a future where voice isn't a key feature in all toys and games devices, experience and leading learning tools in the future. And I think we can all agree. Any vision of the future does incorporate voice technology and particularly when we're coming to our Children. So thank you for listening, just a brief talk and we're always happy to chat more. Um If you reach out to us through our website, thank you.