RESOURCES
For many years now, we’ve been sharing useful resources and media stories related to our research projects. As a leading Science Laboratory in the San Francisco area, it’s important for us to engage with the community and keep them informed about the incredible work and developments they’re helping to support.
RESOURCES
For many years now, we’ve been sharing useful resources and media stories related to our research projects. As a leading Science Laboratory in the San Francisco area, it’s important for us to engage with the community and keep them informed about the incredible work and developments they’re helping to support.
Modern AI speech is perceived as more human-like in older adulthood
Background
Advances in artificial intelligence (AI) lead to an increasing number of encounters with computer-generated (AI) speech in our everyday lives, for example, in smart homes and phones (e.g., Amazon’s Alexa, Apple’ Siri), automated phone services, train announcements, and grocery self-checkouts. Moreover, the tools to generate and manipulate AI-based synthesized speech become increasingly accessible to lay persons. However, research thus far has focused on technology development towards human-like voices and the degree to which AI speech is perceived by young, normal-hearing listeners. How older adults perceive and experience AI speech is unclear and topic of the current study.
The Research
Younger (~30 years) and older (~65 years) adults were recruited for this study. Participants listened to sentences either spoken by humans or computer generated (synthesized) using Google’s AI voices called Wavenet. In one experiment, listeners rated on an 11-point scale ranging from 0 to 10 (where 10 reflected the highest score) how natural they experienced the human and Wavenet voices. In a different experiment, listeners were asked to decide whether a sentence they heard was spoken by a human or computer generated. The first experiment examines how listeners experience the voices, whereas the second experiment tests how well listeners can identify human and AI speech.
Figure: A: Naturalness ratings of sentences spoken by a human or computer generated using modern AI speech (Wavenet). B: Sensitivity to identify whether a sentence was spoken by a human or computer generated.
The Findings
We observed that older adults experience AI speech as more naturalistic compared to younger adults (panel A in the figure). The data further show that older adults identify less well than younger adults. that speech was computer generated (AI speech) compared to spoken by a human (panel B in the figure). The data suggest that older adults find AI speech more human-like than younger adults. This perhaps puts older adults more at risk to be taken advantage of, especially in the current climate where a lot of companies automate voice services. We probably would want to know whether we are talking to a human or a computer, but older adults may be less able to.
Next Steps
Currently, we do not know why older adults are less able to identify AI speech. Self-reported experience with voice-assistive systems did not seem to differ between younger and older adults in our study, but this could be further investigated. We also will follow up with additional studies to examine whether hearing impairment, which is more common among older adults, may explain the current observations. Hence, future research will explore further why older adults are less able to identify AI speech.
Publication
Research Support
NSERC, CRC
Researchers
Björn Herrmann
Key Points
Older adults experience modern AI speech as more human-like than younger adults and are less able to discriminate between human and AI speech.