Olivia Daub’s young son has a strong affinity for “doodidees,” which he enthusiastically requests every morning at 5 a.m., causing a commotion. While many struggle to decipher his words, Daub understands that he craves blueberries, his favorite tiny, dark blue fruit. Daub, an assistant professor at Western University specializing in communication sciences and disorders, highlights the challenge of deciphering toddler speech, particularly for artificial intelligence (AI). She is spearheading a new study aimed at enhancing AI’s ability to comprehend toddler language nuances.
Daub explains that while AI technology like automatic closed captions and virtual assistants have excelled in recognizing adult speech, they often falter in understanding children’s speech accurately. Collaborating with Soodeh Nikan, an assistant professor in electrical and computer engineering at Western University, Daub is focusing on training AI models to grasp the unique speech patterns and errors commonly found in toddlers’ communication.
Nikan emphasizes the necessity of training AI models with child-specific speech data to distinguish between normal speech errors and potential speech disorders accurately. The research methodology involves observing and recording interactions of 30 children, analyzing their speech patterns, and feeding this data into a specialized AI model for training.
The ultimate goal of this study, although in its nascent stage, is to develop an AI model that can aid speech-language pathologists in transcribing and analyzing children’s speech, potentially revolutionizing clinical practices. Daub envisions a future where AI advancements in understanding preschooler speech could enhance tools like closed captioning and accessibility software, granting children greater technological interaction opportunities and societal contributions.