Categorical perception of voice identity?

• ¹ Centre for Cognitive Neuroimaging (CCNi) & Department of psychology, United Kingdom
• ² International Laboratories for Brain, Music & Sound (BRAMS), Université de Montréal, Canada

Although the recognition of identity from an individual voice is crucial for human social interactions, the processes underlying this recognition are still unknown. Contrarily, face recognition has been studied intensively and, it has been shown that the perception of face identity is categorical. Moreover, an fMRI study using morphing between famous faces provided evidence that discrimination between two unique faces occurred in the FFA regardless of physical changes (Rotshtein et al., 2004). The aim of the present study is to test whether the perception of voice identity is categorical. For practical reasons, we used unfamiliar voices of 2 male and 2 female speakers which subjects (n=16) had to learn prior to the experiment. Voice learning was spread over several days until subjects reached a criterion of 75% of recognition of each voice. When the learning criterion was reached, subjects performed two experiments to assess whether the perception of voice identity is categorical: an identity-classification task and an identity-discrimination task. Morphed continua were generated using Straight (Kawahara et al., 1999). One morphed continuum was generated for each gender (cf Figure); initial stimuli were original vowels, and pitch-equalised vowels. For the identity classification task, subjects performed a two alternative forced-choice identity response on each of the 11 levels of morphing. For the discrimination task, subjects were presented with pairs of stimuli from a 1000 steps continuum and performed an adaptative same-different perceptual task. Subjects’ performances in the identity-classification task showed the classical ogival function observed in psychophysical experiments suggesting that voice recognition is partly categorical (cf Figure). Individual data were fitted with a cumulative distribution function; statistical analyses were run on the parameters extracted from the best fit between a theoretical distribution and the data. Results showed that the slope of the sigmoid in the ambiguous region was steeper for female than for male voices (cf Figure). In the discrimination task, we showed that the minimal distance needed to discriminate among two stimuli within an identity was larger for one side of the continua than the other side; the expected result, i.e. a smaller distance for between categories paired stimuli, was not observed in the present experiment. This study is the first one to investigate categorical perception of voice identity. It revealed that the classification of voice identity is sigmoidal, suggesting a categorical perception of voice identity. However, the discrimination task did not allow confirming this hypothesis as it was not easier to discriminate stimuli belonging to two different identities than stimuli within the same identity. Thus, it seems that, contrarily to face recognition, the recognition of voices is not categorical.