Search

Listeners are able to extract a wealth of information from voices: linguistic content, psychological states and speaker characteristics, such as age and sex, can all be decoded from vocal signals. While human vocal communication is uniquely flexible and variable, studies looking at the extraction of speaker characteristics to date have, however, mainly used neutral speech samples. This thesis explores the perception of speaker characteristics from variable vocal signals outside of neutral speech: Nonverbal vocalisations, produced under different levels of volitional control (vowels, volitional laughter, spontaneous laughter and crying) and whispered speech were used to investigate familiar and unfamiliar listeners' abilities to extract and generalise information about speaker characteristics from such variable vocal signals. Experiments 1-2 show that speaker sex perception is impaired for spontaneous vocal signals compared to volitional signals. Experiments 3-4 reveal that speaker identity discrimination is impaired for pairs of spontaneous vocalisations (spontaneous laughter and crying) compared to volitional vocalisations (volitional laughter and vowels), and performance decreases dramatically for pairs requiring generalisation across variable social signals (e.g. vowels versus spontaneous laughter). Experiment 5 shows that while familiarity with a voice can to some extent offset these effects, generalisation is still drastically impaired. Experiment 6 further suggests that familiar listeners are afforded a greater advantage over unfamiliar listeners when extracting identity-related information from voiced vocals signals, compared to whispered signals. Experiments 5-6 thus suggest that a familiarity advantage only generalises to a certain extent for relatively unfamiliar vocal signals (spontaneous laughter, whispered speech). Finally, an fMRI study (Experiment 7) explored the neural underpinnings of the effects described above. This thesis thus shows that 1) the perception of speaker characteristics is affected in a differential manner for different vocalisations and that 2) generalisations of identity-related information across variable vocal signals is only possible to a limited extent - even in familiar listeners.