You may have observed a child attempting to use an Alexa or Google Assistant device and noticed the success rate of those interactions was noticeably lower than when you use it. A big part of that is due to the fact that the speech recognition models in leading voice assistants are tuned toward adult speech patterns, tone, and enunciation. Scanlon points out that children have shorter and thinner vocal cords and have often not learned how to enunciate properly. That means the speech recognition models tuned for adults are looking for the wrong cues when attempting to discern speech from children and the poor results are predictable. 

 

Scanlon began noticing this problem in 2012 and 13 and decided to set out to solve the problem. That has led Soapbox into both the Education Tech and Toys and Games markets. Today we discuss the technology and the applications related to speech recognition for children and even how her company is moving toward developing entirely edge-based solutions with large vocabulary voice recognition that never needs to send data to the cloud. We go deep and wide on this topic that gets too little attention but is likely to shape the experiences of the generation of children born from this point forward.