Olivia Daub’s young son has a fascination with what he calls “doodidees,” eagerly requesting them every morning at 5 a.m. Daub understands that “doodidees” refer to blueberries, a small, dark blue fruit. As an assistant professor at Western University’s school of communication sciences and disorders in London, Ont., Daub emphasizes the challenges of comprehending toddler language, especially for artificial intelligence (AI). She is spearheading a new study focused on enhancing AI’s ability to interpret toddler speech patterns.
Daub highlights the proficiency of automatic speech recognition software in understanding adult speech but notes its struggle in accurately deciphering children’s speech. She mentions instances where toddlers’ requests to virtual assistants like Alexa result in unintended outcomes due to misinterpretation. The ongoing research collaboration with Soodeh Nikan, an assistant professor in Western University’s electrical and computer engineering department, aims to train AI models using toddlers’ distinctive speech patterns and nuances.
Nikan emphasizes the need for AI models to be exposed to examples of toddler speech to differentiate between normal errors and speech disorders effectively. The study involves recording interactions with 30 children to analyze their speech patterns, including common difficulties such as mispronunciation of certain sounds.
The ultimate goal of the research is to develop an AI model that can assist speech language pathologists in analyzing and transcribing children’s speech accurately. Daub envisions potential applications in clinical settings and everyday technology, such as improving closed captioning and voice-assisted software. By enhancing AI’s understanding of preschoolers, there is a possibility of empowering children to engage more creatively with technology and contribute meaningfully to society.