Most people can tell whether a person is trustworthy simply by the way they say “hello.” Something in their voice just gives them away. Now, scientists have figured out what that something is—an advance that could lead to more realistic electronic speech.
“Most of us don’t realize how much identity information is packed into our voices, even in a single word,” says Rupal Patel, a speech scientist at Northeastern University in Boston, who was not involved in the study.
Scientists discovered in 2014 that we convey honesty through our speech. When asked to rank the voices of 32 men saying “hello,” a group of Scottish undergrads overwhelmingly agreed on which seemed the most and least trustworthy. In the new study, the same team attempted to suss out why. Co-led by Phil McAleer and Pascal Belin, cognitive psychologists at the University of Glasgow in the United Kingdom, the researchers created a model voice based on the average acoustical characteristics of the eight voices the 2014 study had rated as most and least trustworthy. The model voice then uttered “hello” twice: once with the tone and pitch of an untrustworthy voice and once with the features of a trustworthy voice. The scientists then used the two versions to generate a continuum of seven low-to-high trustworthy recordings. As a control, they created a second continuum using female voice recordings from the first study.
To find out how people interpreted the synthesized voices, the scientists asked 500 people (146 men, 354 women) to listen online and rank them for trustworthiness. As in the 2014 study, participants consistently agreed on which voices sounded trustworthy.
But this time the scientists figured out exactly what made the difference: Sing-songy voices are seen as more trustworthy, the team reports online this week in PLOS ONE. The most trustworthy hellos “rise at the beginning, drop in the middle, and rise again at the end,” McAleer says. “They are higher in pitch, and the pitch moves around.” In contrast, untrustworthy voices were monotone, with a flat pitch. The voices also tended to rise at the end like a question. The flatter the voice, the more untrustworthy it sounded to the participants. “It was as if we were turning a ‘male voice untrustworthiness’ button on a sound system,” McAleer says. The team also found that some voices fell to the middle of the pitch spectrum. “It’s not a case of trustworthy or not, but that there are degrees of trustworthiness, with one sounding just right.”
It’s not clear yet to the team why a sing-songy voice gives the listener a good impression of the speaker; it may simply be a sign of “a bit more personality,” McAleer says. Many animals, he notes, “warn you off with low-pitched growls,” but use “higher-pitched barks or yelps” to initiate play.
McAleer and his colleagues are now replicating their study with female voices, and they plan to do so across other cultures. Now that they have what the lead author, Pascal Belin, terms a “principled method” for identifying a personality trait in our voices, they also want to apply it to other traits such as dominance, kindness, and attractiveness.
One issue with the new study and the 2014 study is that no one knows how trustworthy the participants actually were, cautions Alexander Todorov, a psychologist at Princeton University, in an email to Science. “We don’t know if [the voices] signal real trustworthiness.”
Still, the study’s results should be of interest to companies working on making speech-generating technologies sound “more humanlike,” Patel says. “McAleer and his colleagues have shown that you can systematically break down and quantify the aspects of a voice that we pay attention to.” And that means that even our electronic devices may soon be making better first impressions.