Voice assistants help keep roads safe for older drivers

Speech data can help predict the risk of future vehicular accidents in older adults, according to a recent Japan study. The use of voice assistants may help capture subtle cognitive impairments in this population.

“[O]lder drivers with accident or near-accident experiences had statistically discernible changes in speech features, implying cognitive impairments,” the researchers said, adding that a “machine learning model using speech features could predict future accident or near-accident experiences with up to 88.3 percent accuracy.”

Sixty older, healthy adults (aged ≥60 years, 55.0 percent women) were enrolled, from whom speech data were collected using voice assistants on modern devices, such as smart speakers and smartphones. Collection was done while performing three tasks: asking for the next day’s weather, booking a movie ticket, and creating a calendar event. Each task involved four open-ended follow-up questions that participants could answer in freeform sentences.

From the speech data, the researchers extracted 84 paralinguistic features, including acoustic and prosodic features. After a mean of 17.3 months, participants were followed-up and asked to answer a questionnaire regarding their driving experiences, including accidents and near-accidents. Machine learning models were then constructed to explore links between speech data and driving experience.

Twenty-six (43.3 percent) participants said that they had had a car accident or near-accident within the previous year, mostly near-misses with other cars or pedestrians. [J Med Internet Res 2021;23:e27667]

Mini-Mental State Examination found no significant difference in cognition between those who reported accidents or near-accidents vs those who did not (p=0.28). The revised Wechsler memory scale, Frontal Assessment Battery, and trail-making test all likewise could not statistically differentiate cognitive performance between the two patient subgroups. Age and sex were likewise comparable.

In contrast, speech pattern proved to be a more useful discriminator. In particular, the mel-frequency cepstral coefficients, jitter, response time, long pauses, speech rate, and the number of phonemes needed to complete a task were all significantly different between elderly adults who experienced accidents or near-accidents vs those who did not.

According to the researchers, the former demonstrated slower speech and less jitter, along with longer pauses and response times, all of which could be indicative of cognitive changes.

A statistical model incorporating speech features predicted future accidents or near-accidents with an accuracy of 81.7 percent as compared to 75.0 percent when using only cognitive assessment data. Combining both types of information achieved the highest accuracy at 88.3 percent.

“Given the increasing demand for car accident prevention involving older adults, we explored the possibility of predicting future accident risks associated with cognitive impairments by using behavioural data that can be collected in everyday life,” the researchers said.

“Although further studies with speech data collected in everyday life and objective data for near-accidents are needed, our study provides the first empirical results suggesting that speech data during interactions with voice assistants in smart speakers and smartphones could help predict future accident risks of older drivers by capturing subtle impairments in cognitive function,” they added.