Identity 2

Person perception from unfamiliar voices​

• There has long been an interest in this: ​

• Pear (1931) ran a radio experiment​

• 9 unknown talkers were presented reading apassage from a Charles Dickens novel​

• Over 4,000 listeners completed a questionnaireabout the different talkers and posted it to thestation.​

• Listeners gave rich and detailed impressions ofthe people they heard.​What kinds of information do peopleperceive in voices?​

• Physiology: Sex, age, height, strength, stateof health​

• Traits/personality: trustworthiness, dominance​

• Psychological/social: social status, accent,language background, sexuality​

• Let’s consider some examples​

Perceiving physical characteristics Talker sex​

• Talker sex​

• Vocal dimorphism in humans is greatly exaggeratedcompared with other apes (Pisanski & Feinberg,2018)​

• Male F0 is on average 75% lower than female; formant spacing(related to timbre) 18% lower.​

• This is due to increases in vocal fold length, and a lowering of thelarynx, during puberty in males.​

• Perception of talker sex:​

• Can be done accurately (e.g. Lass, Hughes, Bowyer, Waters &Bourne, 1976)​

• And very rapidly (within 1.7 glottal cycles: Owren, Berkowitz &Bachorowski, 2007)​

• ​

Perceiving physical characteristics SEX AND ATTRACTIVENESS

• Sex and attractiveness​

• Studies of attractiveness from an evolutionaryperspective implicate F0 and formant frequencies​

• Females prefer male voices with lower pitch, and withreduced formant spacing (“vocal masculinity”)​

• Context-dependent: Low pitch preferred for short-termrelationship, but higher for long-term commitment​

• Complementary patterns for female voices (“vocalfemininity”)​

See Pisanski & Feinberg (2018) for more​

Perceiving physical characteristics​ HEIGHT AND WEIGHT

• Height and Weight​

• Formant measures explain up to 10% of variance in Mand F height and weight (Pisanski et al., 2014)​

• But perceptual accuracy is quite low (Gonzalez, 2003).​

• Listeners may be relying on unhelpful stereotypes ​

• e.g. They might assume low F0 = big, when F0 inreality only accounts for 2% of variation in actualbody size (Pisanski et al., 2014)​

Perceiving physical characteristics​ AGE

• Age​

• Again, estimations can be made but there isevidence for biases:​

• Smaller errors for child voices than young adultand older adult​

• Tendency to over-estimate age for young voices,under-estimate for older ​

• Possible own-age bias: easier to estimate age forvoices around your own age​

• ​

See Moyse, 2014​

Perceiving physical characteristics​ BODY STRENGHT

• Raine et al., (2018), iScience​

• Once again, people can make accurate relative estimates, but there are biases.​

  ​

  Dependent on context: greater perception of strength from roars than from speech in men.​

Perceiving physical characteristics​ SUMMARY

• SUMMARY​

• Listeners can make judgements aboutthe physical characteristics of talkers​

• Percepts of age, size, strength are madereadily and with some accuracy, butevidence for poor use of cues andlistener biases.​

Perceiving traits​

• From Pear to today…​

• Pear asked his listeners to describe speakers based on passages of read speech.​

• But is this how we judge voices in real life?​

• In particular, from an evolutionary point of view, we might rather study our very first impressions of another person (or “thin-slice” judgements)​

• Todorov and colleagues did this with faces: found evidence that consensus on face traits is above chance from 33ms of exposure and doesn’t really improve after 167ms… (Todorov et al., 2009, Social Cognition)​

• McAleer et al. (2014), PLoS ONE​

• Recorded 64 talkers (32M, 32F) saying “hello”: abrief, and socially relevant, utterance.​

• Presented these to over 300 listeners who rated themon 10 scales: Aggressiveness, Attractiveness,Competence, Confidence, Dominance, Femininity,Likeability, Masculinity, Trustworthiness, Warmth​

• Used a Principal Components Analysis (PCA) toreduce the dimensionality of the data.​

What is PCA?​

• PCA stands for Principal Components Analysis​

​

• PCA is a form of dimensionality reduction​

​

• In situations where you’ve got lots of different variablesthat you’ve measured (i.e. many dimensions), PCA canhelp to make the data more manageable andinterpretable.​

​

• For example, it’s much easier to understand how itemsvary on 2 or 3 dimensions rather than 10, 20, or 100s…​

Perceiving traits​ McAleer

• McAleer et al. (2014), PLoS ONE​

​

• Component 1 related to Trustworthiness, Component 2related to Dominance (similar to faces!)​

• Good inter-rater agreement (i.e. a trustworthy-soundingvoice sounds trustworthy to most listeners)​

• Interestingly, attractiveness loads more on Dominancefor male voices, and on Trustworthiness for females​

• How far do these perceptions generalise?​

• McAleer and colleagues have found similar patternsof trait perceptions in further studies:​

• Mahrholz, Belin, McAleer (2018): Generalisation acrossdifferent types of stimulus (words vs sentences; social vs non-social content)​

• Baus et al. (2019): Generalisation across different languages(English Spanish talkers)​

• ​

• This 2-dimensional solution aligns with findings for face firstimpressions, and for social perception more generally​

• e.g. the Stereotype Content Model’s dimensions of warmth andcompetence (Fiske and colleagues).​

• ​

• Acoustic correlates of traits?​

• Tsantani et al. (2016) created high and low F0 versions ofcontext-free words by female and male speakers​

• Listeners chose the most trustworthy or most dominantsounding version out each pair (2AFC task)​

Dominance (High-Low F0) Trustworthiness (High-Low F0)​

​

• The low F0 version was perceived as more trustworthy formale and female speakers, and was perceived as moredominant for male speakers only​

• Low F0 was generally preferred in male voices, while it wasmore context-dependent for females.​

• ​

• Acoustic correlates of traits?​

• Belin, Boehme, McAleer (2017)created a morphed continuumbetween high and lowtrustworthiness “prototypes” fromtheir previous study.​

• Made “caricatures” (i.e. goingbeyond each end of thecontinuum): perceived as moreextreme in trustworthiness​

• Trust associated with higher andmore variable F0​

Perceiving traits​ SUMMARY

Are first impression trait ratings accurate?​

Kramer & Ward (2010)​

• Prepared composite face images offemale UG students who had ratedthemselves highest vs lowest on the Big 5traits (OCEAN – openness is labelled hereas intellect/imagination)​

• From High-Low pairs, participants had topick the image that matched the statemente.g. “More sympathetic”​ . Measured accuracy vs Chance *so above 50%(

Kramer & Ward (2010)​

• One some trials, showed the wholecomposite image (Full) and on others justthe face’s internal features (Internal)​

• Participants could discriminate 4/5 traitsfrom the internal features, and 3/5 from thefull face.​

Are first impression trait ratings of voices accurate?​

• Materials from 93 German speakers​

• Voice recordings (2 spoken sentences)​

• Self-rated Big 5 scores​

• Other-rated Big 5 ratings (first impressions from N=30 listeners)​

• Experiment 1:​

• Made “trait averages” of the top 5 and bottom 5 M and F voices on each trait using the speakers’ self-ratings.​

• N=152 participants listened to pairs and chose the one matching the discrimination statement (e.g. “Which speaker is more likely to be outgoing and sociable?)​

​

• Experiment 2:​

• Made “trait averages” of the top 5 and bottom 5 M and F voices on each trait using previous listeners’ impressions ratings.​

• N=157 new participants listened to pairs and chose the one matching the discrimination statement (e.g. “Which speaker is more likely to be outgoing and sociable?)​

• ​

• Summary​

• Very limited evidence that listeners can perceive the “groundtruth” of personality from trait-averaged voices (Experiment 1).​

• More evidence in favour of stereotypes (Experiment 2): listenersagree with the ratings of other listeners about the perceived traitsof speakers.​

• Experiment 3 replicated above-chance perception of facepersonality ground truths​

• Also found some weak evidence for a correlation in this skill acrossfaces and voices.​

• This in principle could imply a cross-modality mechanism for(accurate) trait perception​

Timecourse of person perception​

• Lavan (2023) found that listeners spontaneously suggestedwords about physical, personality, and social characteristicsto describe unfamiliar voices.​

• ​

• In a later study, Lavan (2024) decided to examine thesedifferent types of impression together​

• ​

• Person characteristics evaluated:​

• “Physical”: Sex, age, health​

• “Trait”: Attractiveness, dominance, trustworthiness​

• “Social”: Educatedness, “poshness”, professionalism​

• Research question: What is the timecourse ofimpression formation for these characteristics?​

• ask​

• 100 voices were each presentedsaying “ahhh” for 25ms, 50ms, 100ms,200ms, 400ms, or 800ms.​

• Rated for the 9 traits​

• Rationale​

• 800ms = should be enough to formimpressions of person characteristics(see “Hello” & McAleer’s findings).​

• 25ms = probably too short?​

​Largely flat time course:High agreement throughout.​

• Physical characteristics areperceived within 25ms ofexposure.​

• Agreement for Sex > Age >Health. ​

• Very different patternfrom physicalcharacteristics!​

• Rising time course:Impressions for socialcharacteristics evolvegradually.​

• Scope for agreement toincrease even after800ms?​

• Different trajectories forthe different traits.​

​

• Trustworthiness andattractivenessimpressions are formedgradually.​

• Although there are funnythings happening at 25ms​

​

• High agreementthroughout for dominance​

• Dominance agreement maybe influences by gendereffects​

Vocal Identity II: Summary​

• Listeners can – and do – make judgements aboutaspects of an unfamiliar person’s identity based on thesound of their voice.​

• These judgements vary in accuracy.​

• Some judgements may not reflect reality (e.g. traits) butoften show good agreement across listeners. ​

• Such “first impressions” may nonetheless be importantas they could affect the decisions listeners make aboutother people.​