Patricia Scanlan: "We Need to Talk about Voice Technology Systems & Children" | Talks at Google
Thank. You thanks for the invitation, they come here. So. I just talk, to you about the company soap box loves to start with we. Do children's. Speech recognition, technologies. So. Alright. So. 2017. And 2018 were like huge, years, for voice, assistance, across, the world it started it was new 2015, when Amazon. First released the election to the US market and it's taking a long time to get to this side of the ocean and. Since then we've seen newcomers come on bit more recent we know that you, know Google came on, the heels of. Alexa. Apple. This year, Facebook. Are talking about a device, by the end of this year they've postponed it but still alexis coming out i'm. What's. Changed, over the last couple years is people are beginning to see the utility of voice assistance, a couple. Of years ago you, know speech, recognition technologies. Were probably working on the 80%. Accuracy. So. Admit they were still quite frustrating, to. Work with so people began just kind of doubted who would ever really take off but myself. Have been working in syria for over 20 years and was kind of a case of once deep learning a vast, volumes, of data and GPUs and all that came together we. Could see the scale of us and we could see the utility um. I spent some time working on IBM a long time ago. IBM, Research and back in 2004, we actually saw a demo of the voice assistant, but, how the idea was somebody would be driving in the car and they, want, to go to a restaurant and the car would ask them you know the system would ask about what. Kind of restaurant. Did you want to go to and, give. Them feedback on real-time feedback on traffic, and direct them to, or ago and it was 2003. I think 2004. So. And, when I was there and I was research assistant at the time I said that was imminent you know this is what they were going to put in cars and you know IRAs standards a year or two and it's actually taken this long to, get real voice assistance, of really utility, and. What's happened there is accuracy it's actually less frustrating. So, tried to a few years ago was extremely, frustrated, got it wrong, too often that's, all wasn't that was getting wrong all the time it's just too often, that was a frustration experience and that's what's changed in recent, years but. It's not just the, voice assistance, right it's. There's any number of these companies that, are doing seriously. Doing speech recognition technologies, right so, I think TechCrunch, late. 2016, said, there was like 29, companies in the US alone do and speech recognition that's not counting Asia it's not counting European companies and it's not anything what's happening 2018, right that's a year and half ago but. All these companies are doing spectacular, things are you're seeing voice, assistance, that you know when we have them first it was theory it was okay Google and was on your personal, device but they taken a leap from the personal device into. The home and that's. Really, interesting from my perspective what's happened there you, know it's. Something we all have to pay close attention to because it's. Just, the tip of the iceberg I mean people think I look this amazing technology what's gonna do for but you, know to my, mind this, is just the start it would be you know even, from our own experiences, of last couple years to. Be on your fridge your TV your car to. Be in a vending machine it's it's, not just that it's a voice assistant it's, actually going to be the stage that. Speech. Technology, voice technology is gonna replace keyboards. Touch swipe. Gesture. They're all these things because it's the natural interface, it's how humans communicate so, it only makes sense that if technology can do it to that stage why. Are we also typing. And clicking and you know we have to do that because the technology wasn't there to, begin with.
But. What happens with that is then we actually start, replaced an interface and voice becomes, the. Norm you have to recognize the fact that you've, put it in the home and children. Will use it I mean, that's just a given it's not about what you want whatever you you volunteer, to take something and put it into the home. It. Will engage children, because number one it's fun you, know what child isn't gonna try it the moment you do it they're gonna do it immediately after you but. It's also easier, for them than clicking and typing cuz a lot times they're even pre-literacy or. They struggle with dexterity. It's actually great interface for children as well and it's already there so the question is how do we handle it what's the utility so in every place that you could see an adult's, using speech, technology, other than dictating, your business documents you're. Going to see that children, will use it to rights especially when you take it outside the office environment on you're. Going to see that their users other, application. Areas we see for it in particular is. And. What we started, will actually still box labs original our mission statement back in 2013. Was more around education, so, if you think about how a child learns to read the. Most effective, way a child learns weed is one-to-one, oral. Dieded reading rights that means and. Helpful, adults working. Alongside a child and if you think they ever watch a parent, or a teacher teaching. A child to read they. Listen, they correct. They prompts, they encourage and the assess write and it's, that one-to-one actually it's, but more than 10 minutes a day but between 10 20 months today is recommended for a child to get to you. Know to accelerate their into it decent, level but, in the world today, literacy. Levels, have stagnated over the last 10 years in the u.s. 60%, of kids at age 8 are not reading at a proficient level and a. Child. That doesn't read at a proficient level at age 8 it's, a key indicator of their future success it's quite a key moment because, it's when most, kids stop learning to really start reading to learn so. If you are. Struggling, before, that it's. Going to get harder and harder to catch up with your peers so one of the ways when we've been looking. This over the years is that if. You. Know the one way to increase the region is one-to-one. Engagement. You're not suddenly gonna find funding to stick more teachers into homes, or suddenly free of parents time or get private tutors what you can do is provide a, scalable. Cost-effective, solution and speech technology, that can actually act as that helpful adult listening, as the child reads the cash sat on the mat and correct, in prompting, and assessing, and do, that personalized, learning journey, for them and the same goes for English language learning or other. Language, learning depend on where from. The. Same thing it's a guided, or a reading so personalized learning personalized, education, is a huge area in the world today. She nobody as a solution really for reading and they can do it for maths and science and geography but the cat do it for reading you can't do it from language learning and, the.
Statistics, About poor, reading is actually masking, a big much. Worse problem for, disadvantaged. Children and immigrant children because those kids are actually two years behind their peers in. Studies. Pisa studies. So. They're the kids who actually would benefit the most from these type of technologies, that are cost effective they're, scalable and all they required as a cheap Android device and means, to means, one of those devices. That's. One of the areas we're focused on another. Area that we've had a huge amount of inbound interested, in social robots it's called or robotics, for formal entertaining I've kept them to move em but also social, robots, engage children it. Gave them in an educational, way or, last hands of kids with learning difficulties, as well, but. Again voice technology to have that meaningful interaction, is really important, there's. More fun stuff like envy or which also plan you know it can be educational but it also can be just pure entertainment and then just pure gaming as. Well I mean where you see voice technology like. For adults keeping in the solutions for children as well makes sense. So. I'm just gonna show you a quick video here and, this, is one of the reasons, why our, focus. In the, last five years has been somewhat different. Just. Play this for you and you'll, understand, what I mean so this is when you stick a voice assistant. Into. The hall and what can go wrong. I. Can't. Find the song too good to her. You. Want to hear a station, for porn detected, porno, read on a hot chick amateur girl. No. Let's, stop so, it's a funny cute video right but it actually indicates. Kind of more, serious. Problems they have with when, you put it by such as home where's the filter I you, know you you can say it's, not designed for kids it's not marked if I didn't put there for kids but kids are gonna use them then we have to face up to that right so. Some of the folks of our company, over the last five years has been accuracy. For children's voices. Systems. Environments, that are fit for kids and data privacy. So. Accuracy why is speech, technology, difficult, for children. Most. Systems, been built with adult data modeling, adult behaviors. Physical. And otherwise, but. Children are physically. Different from adults they're particularly, in the vocal tract their tenor and shorter so, where men's voices might be lower women's or overlapping but higher and then that say a tween. Teenager, is that kind of overlap on women and as they, get younger the. The, physical, differences, get further and further away from, an. Adult. And that happens right at age 12, and the. Younger you get more different it is not actually just physically causes confusion about. What. Actually has been said the behaviors. On the other hand are just wildly. Different kids. Are widely unpredictable. They. You, know they stood her they repeat they shape they whisper they sing like you know you, can you know fail, they. Punctuate, their words no, flap more than and then they're more fluent, older. Siblings. And other children and adults and, that actually is a real problem for end point detection if anybody's ever worked on that way you. Know because they don't just, behave the same way and what you end up doing is seeing that a system. That has been designed for. Adults are using that of data or modeling other behaviors, it's extremely, frustrating for a child and the younger the child the more frustrating, it is has, a negative effect on Brandon, some ways because people associate that technology, they don't a lot. Of people don't process the fact that it's just different voices, today they just think the product isn't working very well and you know I had, a pain in my head for my youngest, son asking me to can. You ask Alexa, you, know when we got a first or you know for. The first one so. What. We've done is spend. The last five years content, trading an scuse me. Pacific. Child Pacific models so what we did was look at deja from. Children's. Young as for, but the data was, conversational. Prompted, read. Spontaneous. And and, built. Models, that were very specific to young children we spent time studying how, children converse, her children speak to technology. The. Variants and not on how do you cope with that and how do you build a system specifically. Designed for children we, collected data in real-world environments so, that meant you know in you. Know where children, are and they're in their home from their schools and. So. You can actually understand, that children don't. You.
Know They don't use. Technology, in a quiet lab like environment and if. You build your. Speech technology, on data that's been checked in a lab like a quiet environment it will only ever work in those environments, so. To move away from that you have to be able to and get away from the headset mics, and get away from all that and then and let children just be natural with how they want to interact with their technology and then it's. More up to us to change our technology, and then the charge but change their behaviors because you. Won't find a five euros gonna modify their behavior as much as an adult will and that's how we've been getting away with a lot with, voice. Technology and the systems that we know that adults sometimes modify, their behaviors to get a response children, are less less. Likely to do so. So. Another issue is is the system fit for children and. Are. They appropriate so, two. Examples I can give you this is that. Sarah. Hook be standards the press, secretary for Donald Trump recently, tweeted that her, three-year-old. Ordered. A Batman. On Amazon, Alexa, because she showed these tried shouted us Batman. At that device three times and suddenly could order it so yeah, I call that somewhat in question, that the UI works like that but. It kind, of did flag up an issue if a. Child uses, a device that has access to purchasing, power why. Is that I needed to do that's. Not appropriate behavior for a child interacting with a device in the home. Similarly. My, daughter. Was she's eight now, she. Was, using my phone and you know I have laughs but she got onto the Siri thingy she was just you know sitting with me and using, Siri to ask questions. Siri. Misunderstood, watch bitch, she didn't but. My my daughter was shocked and showed me the phone with the words printed, to screen and Syria. Replied with her art there's no need for that which is you know very amusing a nice but why. Didn't it recognize that was a child voice and not bring that to screen, so. I think you, know when you consider, the, fact of the you. Know the first video with the child's potentially, accessing, inappropriate material. Inappropriate. Lyrics sounds. Purchasing. Power, you. Know inappropriate words. Potentially. Actually accessing, your contact, list and doing video calling River what. Needs to happen is, and we can work with clients to do is to recognize at the point of. The. Voice interaction whether it's a child whether it's an adult and act appropriately read, the child to a different environment safe, child safe, environment, and not, to the world wide, web where they can do an open search and and such things and that's important, when we put these devices, in the homes and in the cars and in all the places that children will access them, so. Teague's with privacy the really thorny issue of data privacy this is obviously. Such a hot topic in the last few weeks. But. Interested, in the us are way ahead of europe. On this, going. Back to 2012, the. US have couple, laws explicitly. Stated, that voice was, personally. Identifiable, information much. Like video, and images were and therefore. You needed to have explicit. Permission, from the parents, to. Collect data right now explicit. It's very explicit like using their credit card to make a micro purchase or having. The parent. Put. In their email address email sent to the parent parents have to click consent, and then you have to remind them that they consented, like in written in language that they can understand, and it. Was very explicit and weren't. It's right. But. With the advent of. The. These. Voiced assistance. In the home the, FTC. Relaxed. The rules only very recently and said okay we. Understand, that these devices. Are in the home, so, therefore if, it replaces, the voice replaces. What there would have been typed it's. Okay to use it for. Only that purpose and songs it's leashed immediately. That, meant in most people's minds okay we're good to use it in the home, but. What was not considering, was a couple of issues is, you. Know what. Happens on the, back end you, collect. Data from a child is. The. Data deletion immediately, which meant you need a gate for a child or not child data is, the data it, immediately is their data extractor, for that it's the data extractor used for other purposes what's, that data used for so, well, we. Can somewhat, say that we're. Addressing some of these issues it. Has been very unclear, to date exactly. What data's been collected exactly, when the date has been deleted and, what's the data used for and that is because it's an exploding, area and the area is growing so quickly it's, often hard for the laws. And the FTC did, really struggle, with this and kind of came up very short. About. What to do what happens if you've, given permission to your child to use the device and that data's being collected but the child's friend visits sighs you, know what happens to that data is that had been addressed what.
If You just take the device out of the box and put it here you, know is everybody. Being sure that the data is being deleted immediately. And, that's really a 20 issue so the FTC relaxed the room so, a lot of people are now and greywater go and they think we're okay or, we'd the public or who. Have put that device in in the home are unsure when the data's been deleted exactly the. EU in the, GDP, or the. Much-feared gtp, are in 20 May 2018, have, not relaxed the room that, means that, data is not allowed to be collected related less information so you take a device you put it out you put it in your home you plug it in a child uses it where. Is the responsibility. It's. Everybody immediately reacting. Recognizing. That a child has spoken versus an adult and reacting, appropriately so. If people have got the, problem here is that because the FTC, kind, of took, a bit of a back step in the blind eye for a couple of years and then, it got sorted out. The. Idea that that you would do the same it's a little bit less. Clear, one. Reason I would say that is because last year, the Kayla doll was. I recalled, from the market in Germany and, not just was a recalled with the, what, came age was that, parents should destroy, this doll because. The doll is listening, to your children at all times and sending, the voice data to a company. That has links to the US military now, that company was nuanced, which. You know it's a very reputable voice technology company around long time and they have done darfur projects so technically. It was correct but who the, reaction, from Germany, was, so, polar. Opposite to what we've seen already in the FT C in the US that I think it it should give us all pause to. Say how are we doing this are we doing it appropriately and, I. Don't think they're going to see a blind eye so. In that case so we. Have to read envision, how people, are using these technologies apart, what we do is we have a bias companies, and we have solutions, to bid to, help companies, better, comply, with global, data, privacy regulations, so, this is just on the company I mean I can we, can chat more about in a little while but just just on what's happened with the company we you know it starts back in 2013, and 2018 we've, raised a. Couple rounds funding, from privately. And from, the EU as well and. We've got some great coverage from TechCrunch and wired and The Next Web and in recent, months as well. So. We were 10 at the end of the year we're 13 and from this month and then, we're scaling quite rapidly this year like you know I. Can. Leave it there and thank you for listening. Thank. You overview, you talked a lot about your, company and so only take a few minutes talking with you and your journey and to share how, you've gotten to where you button to here and phenomenal. Here to story 2013. And. Pure. Tech built in Ireland. Scaling. Company so so well done on that but, maybe good sleepin if you're your background. I mean the, the journey today right, upload the starter self box labs yeah sure and so. I qualified, as a software engineer, back. In 97. Some. While ago and I spent a couple years working as, a software engineer and worked on building signal processing. Mm. I went back to college, to do a PhD in speech recognition technology which, was you. Know it was a lot of us knew I mean we've been doing speech tech since the seventies properly. But. It was it was so difficult how, we do it now and volumes, we do it with and the approaches, are so different and the the, ambition, was different. But. I spent time I start a new CD when I spent time out in New York in. Columbia University, for then in also an IBM Research in New. York ten Heights research facility, and then when I finish my PhD I, started, in Bell, Labs which, is now Nokia, Bell Labs and I, spent seven years there. Working. On research for them more and more as time went on into the commercialization. Of research innovations. I'm, trying. To I was always kind of pushing for new. Technologies. Or new research, we should do a new product we could bring to market um, I'm. Saying. In. Some ways kind of pitching, ideas and, raising. Value propositions, and trying to get resources and funding to fund projects. Within. The organization, and but. By 2013, identify, the fairly. Significant, gap which, is children's, speech recognition that I. Felt. Like I. Needed. To invest time in and try, and figure out what the problem, was and was something. We need to do myself so I quit. My labs to. Started. To find it supports laughs 2013, and, was that an aha moment or, was this one of these kind of you, build up to see the problem and identified the gap how exactly did you did you can progress this specific, problem and it was all my experience, up, to that point led. Me to the AHA but if you want like this you know I was youth I was working with my daughter was, three at the time if. You. Know downloading. Yep absalom.
Services, And kind, of look at those twelve you know phonics, and get written a reading not stuff like that acutely. Laptop, so that kind. Of data phonics stuff from the reading stuff I was just she, was there, was always MultiChoice. Anything's do it really none language learning was always multi choice because I kind of figure out well big no way of assessing whether she could get this right or not, so she's playing. These beautiful. Games there all done pedagogy. Was great in the world uh but they've no way of assessing and. I, remember asking her she, completed, other levels, or whatever it was and. You. Know I was a food you just finish that and then you know I'd ask her what the same was what the word was good I don't know and I start watching what you do choose just gaming him because she's got around the first time she. Just knew next time I'm through the other one she, had no clue what they were you shouldn't know what the sandwich she's no ability to recall she wasn't even to read the sound or the word. And. To my mind that meant there was you, know why hasn't anybody join things because we know if I could work on speech recognition for so long at that point and I knew where we were going without. Speech, recognition yeah the writing was on the wall everything, we'd be a paste voice assistant, stage. So. It's just baffling to me that nobody that you put in you know, not that can put everything they had they just hadn't. Got. At the same stage you know and. You know it's why I spent, quite a time trying to figure out what the problem was and why. So. It was definitely my experience, coupled with sitting. With my daughter and going ah right there and then realize, inflection, figure tonight. And. In the last five years so I mean you spend a lot of time. Working. With some of the biggest names so me IBM, Bell. Labs you're now in a start-up environment what. Have you taken from from, that time in. The corporate environment so many of us here I were working in big companies we. Like to think we're a start-up in, many ways in Google and but what have you taken and what are you applying in your current, role today that that you you've, learned in, your coursework were Pete's me so, we don't bring products, to market ourselves we. License, our technologies that was one of my learnings probably, for my Bell Labs day it's a you know it's such. A big ecosystem you, know to possibly, I was gonna do everything myself I would put every bit of funding or what a raise would of gone way, too thin, and I, realized that in order to develop the, technology, we. Need and I actually realized slowly. Over time how big a problem was and how difficult a problem was that the. Minute we, started, going to the end-user and we sort of got a B C we, have to invest a huge amount of technology, in atoms. As well I started funding resources. So part, even when we were in Bell Labs we stood you know one of the thinking was and part of the copper things you don't have to do everything you, know you can partner and you can actually just decide them laser. Focus on this problem and then find, partners because you know that's what big companies doing, with perfect it.
Where It make sense, but. Also what what I learned also was that how. To. Talk to corporates and to, know what their problems were and their pain points and realize actually sometimes the. Left hand does know what the right hand is doing so you know what we've discovered sometimes we go to a big organization, get a contact in and top someone they might not think, they have a problem realized a product they have a problem or you. Know, that person is just too busy or they don't, they haven't put it as a price but you might actually talk to a different person in the organization and. This is their pain point you. Know and they will talk to you and then you'll start engaging so I think sometimes people in, you talk to a corporate you get one no thanks you're all right and you think that's it you walk away but. You, know realizing. That very big corporations, it's just not possible to know what every part of the organization is, doing and served as well and. Then realizing it we're, in the organization, try and get contacts into you know what. It's. Definitely relationship, building over the last couple of years and we kind of use a lot of our professional contacts over the last 20 years and to. Help getting to talk to the right people that's, okay. Any, big surprises I mention 2013, to today I'm, sure it's been lots of ups and downs in the journey and but what's kind of surprised your most that. You weren't thinking before you took the leap into starting, a company I. Was. Full, sure at the start that being white if more competitors, right but, I'd been done in my two three years four years into 100 that's why this is. You. Know well this people you. Know really don't fry bigger fish like before, them do this but but we've been very focused, on this year but bi I, think two things happen I surprised. It took as long as it did for the voice assistance should go, but. That's right not that's there's always many factors. For that and. Then, you know in the child speech - I definitely think people are waking up to the fact that that's a problem today again thought. That would have happened earlier but you know that's fine you know we're we're in a great position, good. Good problems or surprises, yeah. And in, terms of yourself, your own time I'm the one of the things that we talk Google's, work-life, balance, trying. To figure out how you prioritize I mean, you've got a lot in your plate you've got the tech side you've got the vision you've got the business and, how. Do you determine your priorities, on a given, week or a given month in terms of what you specifically, are doing versus your team, given. The opportunity, what, you described earlier on is just just, so big I wish. I had said, I know, how I do that I don't we just parked we prioritize, on-the-fly you. Know I mean my job is CEO. And it just. Changes, by the day I mean honestly, you know I'd be. The painter and you know because we just moved into a new office you know where I'll be you know talking. To legal or talking to a can't see or flying. Into San Francisco or you know to New York. Every. Week is different, that's, it, is just you're, just constantly just weighing up stuff in your head, dynamically. Priorities. Well you told us Brody last week suddenly drops because, something else just happened. But. Just having a good team and everybody I think what. Helps priorities, is always you, know having the division, in mind and one, thing we've, never changed is that doesn't. The, voice technology, getting the best possible platform that. Serves into it you, know our different. Verticals. That. Is a system that everybody can use we don't have to customer. For, everybody that all those things have been a focal, point for us so you know we have a number of different clients and they're all looking for different things we.
Actually Say no a lot, that. Helps not. Being afraid to say no because, you know just because you want to see nobody else no other client wants them that's probably not a priority for us like you know, you. Know so we try and build something that that's a useful product to as many people as possible and, then things like that like so I always having that in mind when you're making. Your. Party lists for the week. Pivoting. Into the back. Into the business you presented. A whole lot of logos at the very very start some, of the biggest companies in the world all in this space and. Commercialization. Of voice. Technology where. Do you see. 5-10. Years going what's your perspective on the. Revenue streams that are gonna come. Out of these technologies. For all these companies, I think. A, lot of. Conversation. Now is by winning the home. You. Know I think that's what's a driver for most people I mean you know they're not charging for it but. You know you know. When. You see. What. People are trying to do I think is to draw you into an ecosystem, and. Have that you ecosystem, be as, useful as possible too. And that's where I think there's you know it's not about the voice technology per, se but, you using one device over another and paying for the hardware or maybe maybe, you do you're buying there or maybe your your. Your security your, services, that you you and this are like, who's giving, you the best experience and that's. You, know where, you going I'm what most, of these bigger companies realizes. You. Know you. Know I'm an awful you know fine I've got an iPhone I've got a macbook I've got you know I mean I. Will probably get the home but if they ever fix their you, know those types of things, you. Know so all. Those things I think make you there. You know it's kind of like being a McDonald's, lifer you know that's why I kind of see where people are viewing the utility. And. Revenue streams. But. That's. Not saying to change I do I think I think it's going to be case if your product, will look, less for not having go, to voice technology as an interface you, look like antiquated, technology if, you don't have good and that means it's going to be on you know I mean there's there's cars that are you know partnering, with some pretty big speech to technology, companies you. Know to to. Bring them you know if there's. No way you're the Tesla's got pretty because you know speech recognition. Technologies going you can't bring out a quality product there without us again. You know so over time it's gonna just become more of the norm, you. Know and and everything you sell them yeah majority. Food will be done my voice and, has that evolution, changed, your thinking on, how soapbox approaching the market over the last five years things. Have moved very very fast in in, the, market have, you at the pivot if you had to twist, if you have to change or is your your approach to the same um. We. Started using the word skills in, the last year I think that's changed but it's. Like. So you know but no. Not in some way we always designed it to be and underlying technology, that would serve into education, gaming, you. Know q-tip we always have the utility, box there like which is I think old, wise we talk about voice, control, you know robotics, the air vo all that like toys is a big one as well. What. Changed, a lot for us is over the years that we you, know we, put start putting a stronger stronger focus on taste privacy and data protection and, what that means and we want to be be sure that we relieve in the field not mean we, know how it works or we want to make sure that, we can help other people can we be compliant. No. To my mind it was knowing that was going to be a problem and. You. Better. Okay. And it may be getting a bit more specific, on the offering and so, sitting underneath Tupac Labs there's there's AB, system IP in terms, of development, how. Have you gotten there in terms of like really getting something unique in something different that's, not available the marketplace so the the. The, ideas that tossed the technologies, what, exactly is under the hood um. I think, kind of pointed in a little bit in the talk was about the data, I mean you know we use state of the art deep, learning technologies.
Where. You have a team that's you. Know I think home, we've over eight years experience, in the team and speech recognition technologies, that's actually growing if I got I might have to add to that. You. Know we. Concentrated. On quality, data first. You. Know I recognize that problem, after, my experience in their. Labs and in. Actually. It was Google back, in the. Early 2000s, that started, a really clever of data collection program you, know these group for one one they bring and ask for advice and, you. Know you get a free information call, if, you use voice technology back in the early 2000s, and it, was actually very effective way to its data you. Know this goop, started a long time before anyone else, but. My own experience, I worked on a number of different projects within. The Bell Labs states where lab, like data, versus real world data actually. In IBM as well was, that you know you can connect all the data and world and build a very. Efficient system but if it's not representative of, where you expect, to use. The. Voice technology it. Won't work. You know all those learnings. Led. Us to, to. Do things in a different order though the people were doing yeah. We were concentrating on, quality. Data understanding. The problem, understanding the application, areas and. Then layering on state-of-the-art technology. Or. Cloud-based API, making, it available and, all those things so there's there's, kind of multiple areas of expertise, from, the team and our. Own experience, as well like you know people, who came to them we, you know how we do it how we dress behaviors. All. That there's like as, well as the thousands of hours a day so that's. Good. To hear a lot in there and maybe open that to the floor, anybody. Questions. For chicken. Hi. Thanks, for the talk very interesting I, was. Wondering if you could share some some, information, about the effect. That it has on children too, especially. Under behavior. Right so with. The with. The products that are on the market now it's very often that all, it that is needed is a command or an order basically so. You say like hey Google do, this but. You never say please or thank, you or anything like that so what kind of what kind of impact does that have on the behavior, of a child that they think that that's normal. And. When they ask someone else to sometimes. Something, else there is to someone else like are, they gonna copy. That behavior as well or do they realize that they're not talking to technology. Trends. In Egypt the child I guess as, a parent. You. Know I would. Always advocate, you say please regardless, so I mean it comes from the parent I mean that's which that's the child could be rude to another, human. I think. You teach those behaviors, some, ways the fact that Alexa, will do or a Google home will do it regardless, you.
Know Maybe, that's a little like you know a little add-on or a little like opt-in that you got a do she, won't do it unless you say please. We're. Stop back in Google to address, that issue. My. Question is about the product, is, it more like a classifier, for kids voice or is it, speech. Recognition including kids. Or. For, kids for kids voice versus, is you're trying to predict if the same kids speaking, or children speaking or is it more speech. Recognition including kids trying to understand what they say we do both so. We recognize, adult versus child classification. And then it's also speech recognizer and for the classifier. How. Do you classify how, do you what, is the threshold for being. Voices. Of children voice or adult, voice, so. Traditionally, you know it's, kind of a well-recognized problem, that under, the age of 12, is. When you find the voice started, to differenciate the most so. That depends, on the client, that's, a or you, know coming to us about what their specification. Is would they rather what, would they rather it happen it's not gonna be binary it'll be black and white so. You're not gonna be able to say I've, a cut-off at 13 cuz you're gonna have some 11 year olds they're gonna speak like a 13 year old and 13 world's not speak like 11 year old so you will have that a little. Gray area there. What's, definitely yeah you get it's, quite clear once you start going anyway you. Can. Ask that courtesy. Of. The. Classification. I'm saying this is kids voice or not kids more kids voice and for the age versus, like. I could share more information babies after but it would very much dependent says I don't have older, I think, like, saying if you're gonna say can I recognize a four or five year old from an adults I'll say hundred percent accuracy you're, going to start saying can I recognize a twelve, year old from an adult that gets harder so. It depends on if it's very, easy to separate young children from and also say that. Thank. You listen. Do you have an Alexa at home do you have any other voice assistants at home, no. I don't. Have. A big penis cerium, on home devices though and is. It purely out of a professional, interest that you have it at home or do you actually use it for. Queries. Yeah, it was definitely bought as a research piece. Of equipment to me I think I bought it from the US because I wanted to have it back in 2015, or 2016 and, then I you know I have access to a Google Holmes with me testing that as well. Oh yeah. We use it we definitely, use it like as a yeah each. Of timer's alarms, you. Know playing music is a big one in our house. How. Do you spell this but you know. General. All it's just for fun sometimes like I think I, think the utility, could be better I think we haven't quite seen it yet I think it's quite limited. And sometimes something on Amazon, that could you actually see him working in somewhere like the US where or London, where you got over and you got your deliveries, and you got all that you know the way because you. Know call me an uber that see that's what I want to be able to do like you know we can't do that here in Ireland so I think in Ireland were somewhat limited cuz they've only just released, it. For, Ireland. It's depends on what it's connected and what apps you enable I think. We're someone that was it here I've definitely seen better utility. India, and. From. Your personal experience which one do you find more accurate their likes are than Google. I. Would. Think Google. In the natural language understanding for. Sure. I, would, say I mean, you get way too many I don't know what that means in. Likes. But I think that's an understood problem I think. You know and then the problems vice-versa I think both of them have their strengths that's, what your Linkedin to depends. What you need as well I think I actually people wouldn't, you never really hear ones that's gonna be that strong with the other I, think. When it comes to kids speech I won't say we've. Done our own empirical research on that line. Thank. You, you. Took it you talk about speech recognition of course, but have. You ever is, there, a lot of focus. On the other side so, I mean for me it's it's it's really weird to that. We have this advanced, speech, recognition everywhere, but. The system, that asks you especially in the IRS, and stuff is very clunky stir very, robotic very if you want plasters, so, is there anything going on especially for kids on the. How this is served like this is a fun assistant.
And He doesn't ask. And answer in a way that he might expect somewhat more human-like you know like maybe throwing a joke in or like presenting, in a different way as is completely, these two separate, fields or us the same kind of the same, company. So. The assistant being more of a person. Then then just a piece, of technically, like you know I can relate to that person here's, some silly jokes and I program and for the personality, that kind of thing there's. Nothing really on the market that's blowing us away but, that we know there's a huge amount of companies looking at this. Not. A huge amount of research done was actually funny enough we've we've looked and we've been asked about or recently. Done. Exactly you know what would, a points. Assistant, or a chopped up your child you know what makes more sense because kids use different language they have their concepts, of simple you, can't use a lot of what you would use for an adult but a child because they won't get it and. It's not appropriate sometimes, even the concepts will be just too much I can't they don't get subtleties and, then again it depends on the age group you to information. 47, are you talking about nine twelve you know I mean there's, a lot of very. Little I see very little but. But then again you think which have fought that kind of automated, system, that natural language understanding system, quite new for adults, only getting better others know. What. I've seen it's quite like. What. We've. Seen a lot of people working out. How. He thinks about for the presentation. And because, of the nature of your company I guess you have collector. For access to to a lot of data and, voice. Samples, of little kids so. If you could share like how, do you approach gathering. This data collecting, this or buying, this getting access to it and how do you make sure it is compliant, with all of the privacy, regulations, that you just mentioned yeah so, we. We just do it ourselves we don't don't. Talk too much about how we do it but we've from.
The Get-go and 2013, we've been fully compliant with cop up and made sure I was quite because. I started the company myself likeness, I made sure I was engaged. With you, know privacy, lawyers and understand, this and more so understanding, that we. Didn't really have an obligation to do it in the beginning but you know it. Was one of those things I really told was, going. To be an issue and it should be an issue and it should be addressed so you, know for a small company I don't think it's worth taking risks on something like that so we've done everything ourselves and. And, I'm glad about that as well because. It. Actually was impossible to buy data because, you know there isn't anything equipment, out there not an real-world. Date or nothin text on the mobile device at home you know in a real environment not, an uncontrolled, ways and and to my mind everything else was useless because, you know again like I said if you get, the only Daisy goodbye was fairly. Love like environment and. You know controlled by, an adult or a parent sit there making a child read something that's not real world stuff so. We've. Been very careful to do it ourselves we've. Been careful to be compliant. Globally. And. And. Mm that's that's a lot of money towards like means we can stand over what we've done right. That's a lot of effort I guess - yes. So. What one follow-up question then, do, you don't take into consideration, the. To. Make sure you collect. The samples from you know different countries, different cultures just, to avoid the cultural contra bias yeah we've stated from 170. Countries now and you, know tens. Of thousands were much more than that now yeah. That. Was huge as well actually to be honest because one. Of my you, know I lived in the US for quite a bit you're, trying in different areas and one, of the things I always noted was if you live in New Yorker and you if you drop a pin on the school in New York I'm sure you're not gonna get off, or 20% New York accents you, know so why. Just build. A system with us accents, like you know and that was really key that, if Assistants going to work everywhere you the, world just working you. Know you've got such you know flow of people whatever and and, to be honest like deep learning and all those technologies. At the add ones that has allowed us to be able to add multiple. Variations. In pronunciation, to. The system where it's you know when I started this back today as we were you, have like you know Midwest, you probably remodel from Midwestern, us, you know well British look you know and even even. Aren't that you probably have to break learnt into different dialects and accents and stuff like that because the systems couldn't cope with the variation, where we can now. Hey. Thank. You for the talk really interesting, I actually, got interested in the topic because of that week and, talk. About that as well on the future of search so, it had a lot of similar. Implications. And news of what happens if the child is in the room and also, what does that mean for advertising, in voice search do you have any take on the combination, of advertising. In, voice search in the future and children being in the room because there's a lot of regulations. And problems without you yeah. I think be much like I said about the appropriateness, like you know I mean -. You. Know. It's. Probably you know for advertisers, I think you don't waste time. Advertising. To, children like, us they're, not they're not the buyers so.
You Know I mean as much as you should you know you athle, need to be able to make it an appropriate experience, for a child. You. Also, you know commercially. Shouldn't. Be wasting mattress. Isin dollars into. A chart so I think you, know okay and a simple I don't charge classifiers, for different age groups as well is it's appropriate, you know even on apps, now these days you know there's. Quite strict regulations, are not regulations for my guidelines as well about what you should advertise to children so one some are free, you. Can slot in static, ads but it should be a an, add-on Lego or Barbie, they shouldn't be an addressable, inappropriately. And. I think that's just you. Know it's in sense really as well as just you know I mean that there's one side that's regulations, gonna force us all to do these things correctly you, know it's about time, but. On the other side there's definitely, just just, good sense of I so you know you know as much as you target, your ads to different. Demographic, demographics. Anything. Ok, any more one more, question. I'll. Take the last one and. You G people will start running at five ten to get there our next meetings I'm guessing a lot of people here will be doing that and we've, talked a lot about the hard work to date, the problems you're solving the things are going after let's. Go out in ten years I mean your hope for, soapbox, labs where. You'll be what you'll be doing yeah. Oh. Yeah the, future yeah, we you know initially. Are for the next few years anyway we've got a big focus on multilingual, I think. That's a huge thing you know I mean we can't just continue to expect you know it's been such a time to get anything, quality, out there in English but we've learned so much we've got we've. Got our pipelines, we've got our processes, we can accelerate now add in languages and that's what we raise do you funding for actually we raised quite considerable funding at the end of last year. To. Take up for multilingual so that's gonna be a massive, you.
Know Focus of the company and, eventually, there's. Lots of different variations of speech all types of voice technology where voice comes into it and it's kids that's where we'll be and. We're already building up expertise a number of different areas you know doing, more a natural language understanding as, well around children things like that that it's we, think and we listen a lot to what our clients are saying to us what they want. To. Be able to bring to market like you know so you know as much as we will do certain amounts of us at, the moment but, we're listening all the time to what they want and you know but, actually good it's too much detail you know we're. Responding to that and. Right now we have a great position in market because we have you, know quite, a lead because. We focus we focused, solely on that we haven't diversified, over the last five years on kids but, there's so many opportunities globally, there's, so many opportunities in this like self yeah we'll just keep of it yeah well. Great space great company and thank you very much for taking, the time coming to talk to us here sharing, I mean your journey sharing the soapbox app journey, best, wishes for the future I mean I, know pretty much every listener is probably excited about where voice is going where. The technology's, going and it's great to see an Irish company indigenous, company doing, real tech work here put, an ambition, to go, global as well so thank. You very much. You. You.