Patricia Scanlon: ‘We have built a dataset of 25,000 children’s voices’

3 Oct 2017

Patricia Scanlon on stage at Inspirefest 2017. Image: Conor McCabe Photography

Soapbox Labs is on a mission to help children be understood in a world powered by voice.

Dr Patricia Scanlon, founder and CEO of Soapbox Labs, has a point. Forbes declared 2017 ‘the year of the voice search’. The influential Consumer Electronics Show declared 2017 ‘the year of the voice interface’. And visionary analyst Mary Meeker predicts that by 2020, half of all web searches by 2020 will be done by voice.

And yet, under-12s have no voice – or at least are seldom understood – by the machines.

‘Deep learning is a huge factor. It’s the evolution of the technology, so stay tuned’

And that’s where Soapbox Labs comes in. The company is building speech technology specifically for young children that will help with learning, literacy and much more.

Deep learning for little voices

Soapbox Labs uses deep neural net-based speech-recognition technology to assess children’s speech in real-world noise environments. This solution easily integrates into third-party apps and web services, enabling a huge range of applications, from childhood literacy and language learning through to voice control in gaming.

Scanlon was, appropriately enough, telling her story at Inspirefest 2017 in July – Soapbox Labs credits the Astia Showcase at Inspirefest 2016 as key to it landing a €1.2m seed investment from Astia, Elkstone and Enterprise Ireland as well as private investors.

According to Scanlon, speech technology, after many stops and false starts, is now in the ascendency. “Things have changed and we have now reached 95pc accuracy. Deep learning is a huge factor. It’s the evolution of the technology, so stay tuned.”

Scanlon pointed out that more than 36m voice devices, such as Amazon’s Echo, were sold in the US last year.

But, for Scanlon, who was inspired to focus on speech recognition by her young daughter after a 20-year career in technology with IBM and Bell Labs, speech technology tends to be designed for adults, not kids.

“It was clear to me four years ago when my daughter was three that speech recognition doesn’t work for kids. How do you get speech tech into the hands of kids who could use it? Imagine a child using an iPad or a cheap smartphone to help them with their reading or to learn English.

“If a child is pre-literate and can’t read, why shouldn’t there be a natural interface? And it makes sense it should be used for children.

“Voice-enabling technology for kids is important but it is a huge problem. Adults are not children; children have shorter, thinner vocal tracks. The reality is, speech technology for kids nosedives from the age of 12 downwards.”

And this is the problem that Soapbox Labs aims to address.

Scanlon said the world is changing. “Screens are disappearing and it is vital that kids can use their voices to explore.”

To this end, Soapbox licenses its technology to third-party developers and platforms to voice-enable any technology product for reading, robotics, language learning, smart toys and more.

“We have build a dataset of 25,000 children’s voices in 150 countries. These are real children from real places.

“If you want to leverage deep learning and the improvements in accuracy, you need to leverage massive amounts of data.

“We have developed a cloud platform that works with real-world devices and allows a child to simply interact with devices. With this, we can do reading and language assessments, enable progress tracking and we are enabling biometrics to identify between different children as well.”

With that depth of technology, expect Soapbox Labs to give voice to entirely new generations of learners for many years to come.

Inspirefest is Silicon Republic’s international event connecting sci-tech professionals passionate about the future of STEM. Super Early Bird Tickets for Inspirefest 2018 are on sale now!

John Kennedy is a journalist who served as editor of Silicon Republic for 17 years