Patricia Scanlon, CEO of Soapbox Labs, on Speech Recognition That Actually Works for Kids


Okay voiced by listeners. I've been looking forward to this Patricia. SCANLAN IS CEO of Soapbox Labs Patricia. Thanks so much for joining me today. Thanks Oh man okay so to get started. I know a lot of people know about soapbox labs but I think a lot of people do not and so why don't you this sort of introduce yourself. Elf what you do with the company and what is so box labs do at a high level short. I'm the founder and CEO Soapbox knobs on a very high level. We are voice his technology for kids so we power third party APPS web services products that want to voice enabled for children and this is is important because Asr's automated speech. Recognition are generally tuned to adult voices adult speech patterns in Africa children correct. Yeah historically cleo always has been it's it's pheno people often told us I know kids as us like just another accent. We've just throw of audio into you into the models have been built frauds. Motley novels behaviors. Just tune it to a kid's voice on it just does not work. Historically clean never worked on on people are still trying to unfortunately no. That's why we pretty much standalone Nariaki for speech recognition for children Tick because we had laser. These are focused on this problem for over six years right. So how did you come to recognize. This is a problem we we can go back act to the fact that obviously your your time at UCD you're really focused in speech recognition. But a when was it that it came to you that hey kids are is different in the way the technology handles speech wreck. I had my own house. Basically my daughter was three at the time mm-hmm and I was observing her interacting with technology. You know she was it was. It was kind of twenty thirteen and with the dawn of Oh absolutely ipod everybody. There was at millionaires. All over the face of people really invested in making a lot of cool apps up through the in education that age group as well. I was observing her. Interacting with a reading up those teach ner emergent stage reading where you're teaching them sounds and blend them and the Dakota his watching in how a really cool piece of technology that actually it's been designed with pedagogy experts in university collaborations beautiful graphics and really it was a great a product but a reading light on the ability to assess her pronunciation. Recall what was trying to teach her so seem. I'd spent a thought point joint. I've been in the area of speech recognition for like thirteen over thirteen fourteen years. I'd always worked on speech recognition to me just seem so obvious that we should be using voice these technology to assess a child's pronunciation recall and I don't know when you're read or learning language and it just really struck me that wow there's nothing thinned out there that doses ANA level of accuracy for children's voices so I started to investigate researcher for many years. Why is this a problem? You know why nobody managed to solve this. Given the fact was twenty. Thirteen right to think that it was everywhere. The technology space was gaining gaining traction and gain not quite becoming accurate Still Good Way to go with twenty thirteen while I was working in the space and seeing the leaps we're making in Adel speech recognition and then looking at this neglected area of children's speech I'm realizing there's a huge gap here a huge opportunity. You know from an entrepreneur to be addressed To be able to give children a voice to be able to let them be hard weather they were reading or learning the language or gives playing with a toy or game. I know seen how different a child's speech is. You're talking from age three. You're very. The child is very physically. Three different from Exposes a good way to describe Indus- their vocal tracks are thinner. VOCA trucks are shorter on vocals. Walter smaller. I'm what happens in not as the of the signal. The voice signal actually resides in different parts of spectrum. were certainly former frequencies concede to Then the speech behaviors are very different than adults rights of pink or five year old seven year old or nine year old. How they speak? Take the patterns of speech Elongate Over a nun seeds there. They'll sing the whisperer. Spur the But they also don't tend to follow language either so this whole series of aspects to this stuff should inform Sola should be should have been more obvious feeless like It will fall apart. Adults Systems are trained on adult voiced as adults speech behaviors With full apart with kids on the younger you get worse spouse was in my learning. By absorbing observing in my own daughter I mentioned to scratch heckle. Why is this knocking solved in on quite a bit of Exact problem

Coming up next