If you’ve ever used Siri — or any of the other dozens of virtual assistant style apps out there — you probably know firsthand how useful speech recognition technology can be.
It wasn’t always like this, of course — speech recognition has come a long way since the first speech recognition products with practical applications were launched in the 1980’s and 90’s. But in all those years of development, there’s been one key oversight that researchers are only recently beginning to tap into: Speech recognition software has historically struggled quite a bit to recognize children’s speech, especially when compared with its abilities to recognize speech produced by adults.
In a recent article for TechCrunch, Margery Mayer, former president of education at Scholastic, broke down some of the recent advancements in speech recognition technology that have made it possible for such systems to finally recognize and understand speech produced by children. The technology still has a lot of room for growth, but its potential applications could be a boon for software developers and educators, Mayer argues.
“Speech recognition has advanced to the point where it can recognize and process children’s speech and account for differences in accents or dialects,” she writes. “Thanks to the high accuracy and performance of this technology, elementary school educators can now rely on it to help them gauge students’ progress with more regularity and offer more personalized approaches to their instruction.”
In her piece, Mayer breaks down the efforts she and her team at Scholastic made to utilize speech recognition software in 1999, in order to teach young children how to read and assess their reading abilities. While they’d wanted to create a program that children could read aloud to and receive automatic feedback on their reading abilities, the software was simply incapable of accurately recognizing and assessing the way children spoke.
Children, of course, have much more highly pitched voices than most adults, due to the fact that their vocal cords are shorter and thinner than most adults. Speech recognition systems, which are largely trained using data from adult speakers, tend to perform less well with children speech due to the fact that children simply speak at a slightly different frequency than the one the speech recognition systems were trained on.
Mayer notes, however, that modern educators and technological innovators are likely to be more successful in attempting to incorporate speech recognition technology into an educational setting, as companies like SoapBox Labs have developed speech recognition technology that is much more successful in deciphering the speech produced by young children. Now that speech recognition software is better developed for children, Mayer argues that it can be incorporated into reading lessons by providing young students corrective feedback on their reading abilities instantaneously.
“In a few years, this technology will be part of every classroom instruction, accelerating the reading — and math and language — skills of young students,” she writes. “Educators will find it enables them to be more strategic in their instruction.”
