Abstract
An estimated 3.25 billion voice assistants (VAs) are in homes around the world, but these devices are not always able to recognize and respond to children’s speech. To inform the design of VAs that support kids, we report on a lab study where 28 5- to 10-year-old participants interacted with a commercial VA to: 1) attempt to execute common VA-supported requests (such as setting an alarm), 2) recite a set of such scripts verbatim, and 3) engage in unstructured conversation. We find that devices only respond appropriately to the content of children’s speech half of the time. Frequency of appropriate responses increased with children’s age and as their discourse became more standardized. Based on themes in participants’ speech, we identify design opportunities in child-VA interaction, such as exploring a topic or responding to a conversational bid. In addition to our empirical findings, we contribute a structured corpus of children’s speech. • The VA speech recognition system we tested (the Amazon Echo) transcribed children’s speech correctly more than 84% of the time. • When children recited pre-scripted speech, there were age-dependent differences in transcription accuracy, with accuracy increasing with age. But when children were free to formulate their own (unscripted) statements, these differences disappeared. • The VA only responded correctly to children’s utterances half the time. • Children’s speech displayed structure that was common across participants; the VA was more likely to respond appropriately to speech that conformed to common structures than it was to speech that reflected unique structure. Children were more likely to use a common structure with some tasks (such as requesting a song) than with others (such as setting an alarm). • When addressing the VA on any topic of their choosing, children’s speech reflected a thematic set of goals. The most common goal was attempting to build a relationship with the device.
Accepted Version (
Free)
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have