Highlights đź’ˇ:

Source: @yang2020

HinzugefĂĽgt am 2022-11-06

â–  first gender-balanced study of charismatic speech (p. 2)

â–  using characteristics of charismatic speech in textto-speech synthesis can make a computer-generated voice more trustworthy [9] (p. 2)

■ Cullen et al. [12] crowd-sourced charisma ratings on an Irish politician’s speech and built automatic systems to detect charisma (p. 2)

â–  we selected 30 male and 30 female speech clips from Youtube and pilot tested these to balance charismatic, boring, and neutral groups for each gender (p. 3)

â–  avoided voice clips from celebrities (p. 3)

â–  prepared talks, educational lectures, and interviews, and were each 20 seconds long (p. 3)

â–  Amazon Mechanical Turk (MTurk) to collect ratings for the 60 voice clips from 15-20 raters each (p. 3)

â–  demographic questions, including birth gender, gender preference, and level of education, and completed the Ten Item Personality Inventory (TIPI) (p. 3)

â–  each worker was instructed to rate 10 clips on charisma and 17 other traits: boringness, coldness, confidence, eloquence, enthusiasm, extroversion, fluency, intelligence, introversion, liveliness, ordinariness, persuasiveness, reasonableness, sincerity, trustworthiness, uncertainty and weakness (p. 3)

■ How do raters define charisma in terms of other speaker traits? Does the genre of the recording influence charisma ratings? Does speaker gender influence raters’ charisma ratings or ratings on other speaker traits? What are the acoustic-prosodic and lexical properties of speech rated as charismatic? Do raters’ demographic information and personality characteristics influence their ratings? Does raters’ own speech correlate with their charisma ratings or their demographics/personality? (p. 3)

■ Pearson’s correlation, Krippendorff’s alpha, and paired t-tests (p. 3)

â–  average charisma rating of 3.20 in range 1 to 5 (p. 3)

â–  60 clips we collected, 14 are interviews, 19 are educational lectures, and the other 27 are talks to more general audiences (p. 3)

â–  interviews are less charismatic than both educational lectures (p = 0.009) and talks (p < 0.001) (p. 4)

â–  Male speakers were rated as less (p. 4)

â–  we extracted 12 acoustic-prosodic features (p. 4)

â–  maximum, minimum, mean, and standard deviation of pitch and intensity, harmonics-to-noise ratio (HNR), jitter, shimmer, and speaking rate measured by the number of syllables per second (p. 4)

â–  normalized the mean pitch of males by 119 Hz with standard deviation 19 Hz and females by 210 Hz with standard deviation 27 Hz using mean values for American English speakers (p. 4)

â–  mean intensity (p = 0.013), mean pitch (p = 0.002), speaking rate (p = 0.001), and variance in pitch (p < 0.001) were all positively correlated with charisma (p. 4)

â–  louder, higher, faster, and with greater fluctuation (p. 4)

■ So, not all acoustic-prosodic features of charisma that were found for all speakers were present within different genders. The mean intensity was only correlated with females’ charismatic speech, while mean pitch and speaking rate were only correlated with males’ charismatic speech. (p. 4)