The effect of voice identity training on speech understanding in noise

Poster Session E - Monday, April 15, 2024, 2:30 – 4:30 pm EDT, Sheraton Hall ABC

Joseph Rovetti¹ (jrovetti@uwo.ca), Aditi Nayak¹, Ingrid Johnsrude¹; ¹Western University

Noisy environments make it difficult to understand speech, but familiar voices such as those of a spouse are more intelligible than novel voices in noise. Listeners can even be trained to become familiar with voices in the lab. However, it is unclear what sort of training is necessary to achieve this familiar voice benefit. We hypothesized that the more listeners were trained to associate the talkers’ voices with unique personhood, the more intelligible these voices would become in noise. To test this hypothesis, we recruited 240 participants online via Prolific and assigned them to three groups. In the Control group, participants listened to one hour of sentences spoken by two unfamiliar talkers and completed a semantic judgement task. In the Identity group, participants instead judged which of two possible talkers had spoken each sentence. Finally, in the Bio group, they completed the same task as the Identity group but were also given photos and biographies for the talkers. After training, participants in all groups completed a speech intelligibility task with sentences spoken by trained and novel voices, presented against two levels of background noise. We found that trained voices were more intelligible than novel voices overall. The benefit from training was greater at the more favourable background noise level. However, this benefit did not vary between groups. This suggests, despite our hypothesis, that the familiar voice benefit is not supported by talker personhood. More general mechanisms, arising from simple exposure to or engagement with the voices, must instead be considered.

Topic Area: PERCEPTION & ACTION: Audition