Researchers in Australia have embarked on a project that could ultimately help voice interfaces to interact with Australian children.
As ABC Radio Sydney reports, the project entails the collection of speech data from Australian children aged three to 12. It’s dubbed “AusKidTalk”, and is being led by Dr. Beena Ahmed, a senior lecturer at the University of New South Wales’ School of Electrical Engineering and Telecommunications. It also features researchers from the University of Sydney, the University of Western Sydney, the University of Melbourne, and Macquarie University.
Participants are asked to perform activities including reading sentences aloud and also reading out nonsense words concocted by the researchers. The kids sit through hour-long sessions involving iPad games in a recording studio.
Dr. Ahmed’s team is aiming to record the voices of 750 children by June of next year. Currently, there are only 14 databases of children’s speech in the world, and none of them feature Australian voices.
Children’s speech can vary considerably from that of adults, and the performance of commercial voice-based AI assistants like Siri and Google Assistant can vary based on the accents of end users. The AusKitTalk database could help developers of such systems to better train their AI solutions for interacting with Australian kids.
In addition to potentially helping to improve interaction with AI assistants on various devices, the AusKidTalk researchers are hoping to perform other kinds of analyses on things like how children’s speech patterns change as they grow, and how emotion affects their speech.
Source: ABC Radio Sydney
December 23, 2020 – by Alex Perala