Sensory seeks to make it easier for developers to create speech recognition applications for children. To this end, the company has released new speech recognition models that have been specially trained to recognize the unique voice patterns of children.
In this context, Sensory found that many speech recognition algorithms do not work as well in minors as they do in adults, mainly due to the fact that children’s sentence structure and vocabulary are different from those of their parents. The new Sensory models are designed to address this problem and provide developers with a reliable solution for younger audiences. Sensory spent several years collecting and analyzing speech samples from children to achieve this goal and eventually delivered a product with a 33 percent lower word error rate for children when competing with an algorithm developed for adults.
The new recognition is compatible with the TrulyHandsfree and TrulyNatural solutions from Sensory and is available immediately via the Sensory VoiceHub developer portal. TrulyHandsfree allows the platform to recognize common phrases while TrulyNatural fills in the rest of the vocabulary. TrulyNatural can also run locally on a device, which is noteworthy as the resulting solutions do not require data sharing and therefore the privacy of minors is respected.
“We challenged the team to develop private and accurate recognition for children’s speech, and they did it,” said Todd Mozer, CEO of Sensory. “This opens up new and fun voice-activated products for children of all ages.”
Sensory believes the kid-friendly voice technology can be used in educational apps and enable voice control in toys and wearables for a younger market. The new language model is already catching on with developers and is now being used in integrated circuits from Generalplus Technology. Sensory itself recently updated the TrulyHandsfree engine to include support for Apple’s Siri voice assistant.
(Originally published on FindBiometrics)