Kai Alexander Hackney, Lucas Guarenti Zangari, Jhonathan Sora-Cardenas, Emmanuel Munoz, Sterling R. Kalogeras, Betsy DiSalvo, Pedro Guillermo Feijóo-García
8 min
Abstract
To foster effective human-agent interactions, designers need to identify characteristics that could affect how agents are perceived and accepted, and to what extent they could impact rapport-building. Aiming to explore the role of user-agent synchrony, we assessed 388 participants to determine whether they could perceive personality traits from four artificial voices we selected and adapted from human samples, considering gender (male or female) and personality (introvert or extrovert) as grouping factors. Our findings suggest that participants were able to significantly differentiate female agents by personality, while male agents were not consistently distinguished. We also observed evidence of personality synchrony, where participants tended to perceive the first agent as more similar to their own personality, with this effect driven mainly by male participants, especially toward male agents. This paper contributes findings and insights to consider the interplay of user-agent personality and gender synchrony in the design of human-agent interactions.
Alex: Right—like the waver in someone's voice when they're nervous. And the study found these clues worked better for female voices?
Sam: Yes. Listeners could reliably distinguish introverted from extroverted traits in the female-generated voices, but not in the male ones. This suggests female voices transmit personality signals more distinctly, perhaps due to clearer differences in pitch or energy. For male voices, those signals blend together, making them seem neutral regardless.
Alex: With such a targeted setup, how exactly did they pick those real voice samples to feed into the AI generator?
Sam: They ran a separate online collection first, recruiting hundreds of adults through a crowdsourcing site. Only 11 usable recordings provided the voice clips needed, after people self-reported their traits on that survey. From those, the team hand-picked four distinct ones: introverted and extroverted for both male and female speakers. This small but focused set let them capture real sound patterns tied to outgoingness.
Alex: With only 11 originals, doesn't that limit how general the findings are?
Sam: The paper acknowledges the small preliminary sample as a starting point, suggesting larger collections could strengthen it. Still, listeners distinguished extroversion clearly in the female-generated voices versus the male ones, pointing to robust gender differences in how sound traits signal personality. This hints at acoustic factors—like pitch range—being more distinct in female voices.
Alex: So with those AI voices ready, how did they actually test if listeners picked up the personality differences?
Sam: They recruited 429 adults online, but 388 finished the full survey. Everyone took the same personality survey for themselves first, then heard two voices in a row and rated each one's outgoingness using that same survey. The setup was within-subjects, meaning each person compared both an outgoing and a reserved voice directly, which helps spot differences clearly. They split people randomly into four groups: two groups heard male voices first—one order outgoing then reserved, the other reversed—and the female groups did the same.
Alex: Okay, so direct back-to-back listening to make contrasts obvious. What script did these voices read during the test?
Sam: The AI voices spoke a passage from an article on job burnout—explaining what it is, why it happens, and ways to fix it. This neutral topic let sound traits like energy or reserve stand out without distracting words. After each voice, participants scored its personality on the survey, then guessed its age, ethnicity, and gender.
Alex: With that many listeners doing paired comparisons, how did the results show female voices signaled personality better?
Sam: They calculated outgoingness scores from the survey responses for each participant and the two voices they heard. Then they checked correlations between those scores, especially looking at gender matches—like how men rated male voices. Listeners reliably separated outgoing from reserved in female voices, but male voices blurred together. Plus, people nudged the first voice's traits toward their own, strongest for males with male agents.
Alex: So that pairing and scoring setup highlights why female voices transmit those cues more sharply.
Alex: Okay, but to dig into those patterns, how did they measure if people actually distinguished the personalities in the voices?
Sam: They looked at how people's ratings lined up between the introverted and extroverted voices each heard. If listeners consistently gave low outgoingness scores to the reserved one and high to the energetic one, that shows they picked up the difference. For female voices, the link was strong and reliable. Male voices showed no meaningful pattern.
Alex: So they checked if hearing one first changed perceptions?
Sam: Yes, they split groups by who heard introverted or extroverted first and compared average ratings. For first-exposure to female voices, groups differed clearly, supporting their idea that even one voice reveals personality. Male first-voice groups showed no gap.
Sam: Overall, people nudged the first voice's traits toward their own—a modest but notable match. This held strongest for men rating male voices, suggesting they project personality when cues are faint. The paper cautions uncontrolled factors, like a southern accent in one male sample making it seem more outgoing than planned.
Alex: So pulling this together, people only projected their own traits onto the first voice, not the second. That suggests the effect depends on having little info at the start.
Sam: Exactly. Once listeners had a reference from the first agent, their ratings for the second one anchored to that baseline instead of their own personality. No self-projection showed up there—the paper suggests this happens only when agent details are sparse early on.
Alex: With those patterns in mind, what limits should we keep in the back of our heads here?
Sam: The study flags a few. Self-reported traits from the survey might not match actual vocal signals, especially for males where cues were weak. The voice samples were just 11 clips, hand-picked without full demographic controls like accents. Participants skewed young, college-aged, and less diverse, so findings may not extend widely. Real interactions add complexities beyond paired listening.
Alex: Right, so those choices make sense for a focused test, but call for caution on broader claims.
Sam: Still, it points to meaningful gender differences in voice design for agents. Future steps could use larger, varied samples and precise trait-configuring to test in real settings like education or mental health support. Tailoring early voices—say, matching gender and energy—might boost trust.
Alex: That feels like a practical nudge for designers without overpromising. Well said, Sam. That's our look at voice, personality, and gender in human-agent interactions. Thanks for joining ResearchPod.