Perceived credibility of native and non-native accents during voice-AI interactions

When a human speaks with a non-native accent, it can affect the listener's interpretation of the message. The current study investigated how listeners respond when AI speaks with a non-native accent, and whether listeners’ usage of voice-AI affects those responses. Participants, who were L1 English speakers from the U.S., listened to statements produced by neural text-to-speech in Mexican Spanish-accented English (Experiment 1) or European Spanish- and French-accented English (Experiment 2), as well as American English. They rated the credibility of the information content as well as the voice assistant itself. Results of Experiment 1 reveal no overall difference in ratings for the two accents, a finding which differs from previous results with human voices. Frequent users of voice-AI also gave more positive ratings overall, and this effect was largely due to higher ratings for Mexican Spanish-accented voice assistants. Results of Experiment 2 reveal that a distinctive pattern of responses to the European French accent, compared to the other two accents, and frequent users of voice-AI also produced a distinctive pattern of results, compared to infrequent users. We interpret these findings in terms of anthropomorphism, namely the extent to which frequent versus infrequent users of voice-AI are willing to attribute human characteristics to a non-human entity.