Fig 1.
Acoustic description and alarm ratings of 6 distinctive types of screams.
(a) Spectrograms of example stimuli for each scream type, including a neutral scream of an intense vocalization of the vowel /a/. (b) Average modulation power spectrum (MPS) for each scream type (upper row) as the numeric (middle row) and statistical difference (lower row; n = 60 per scream type; n = 420 screams in total) of each scream type from neutral screams. (c) Using a support vector machine (SVM) on the acoustic features of each scream allowed us to separate each scream type from other screams with high accuracy (e.g., low level of off-diagonal misclassifications). (d) To test whether screams share acoustic information with common nonverbal affect bursts, we trained the SVM either on the screams (left) or on the nonverbal affect bursts (right) and tested the classifier on the other type of affective vocalizations; in both cases, the SVM classifier failed to show superior performance (e.g., high level of off-diagonal misclassifications). (e) Alarm level ratings (n = 23) confirmed the differential level of alarm of screams that seem to be of a low alarming nature (non-alarm screams: pleasure, sadness, joy) and a high alarming nature (alarm screams: pain, fear, anger). (f) Response times for the alarm ratings as performed in a non-speeded task by moving a slider on a visual analog scale to the appropriate alarm level for each scream. Numerical data underlying the plots (c–f) can be found in S1 Data. ang, anger; fea, fear; neu, neutral; pai, pain; ple, pleasure; sad, sadness; sem, standard error of the mean; test, testing data; train, training data.
Fig 2.
Perceptual decision-making on perceived scream calls.
(a–d) RTs and accuracy level for (a) categorizing the 7 types of screams in a 7AFC task (top left) in experiment 2 (n = 33). (b) During misclassification of screams (top right, off-diagonal), participants significantly used more categories from alarm screams than from non-alarm screams (bottom right). (c) Combining the hit rate and the false alarm rate results in the d′ measure of sensitivity to detect a certain scream type (bottom left). (d) The false alarm rate is also reported separately for alarm (red) and non-alarm screams (blue). (e–h) The experiment 2 was replicated with an independent sample of n = 29 participants but using a new selection of 84 scream stimuli. The data were analyzed identically to the data reported in (a–d). (i–m) RT (i) and accuracy level (k) for discriminating scream calls (experiment 3, n = 35) from neutral screams (first set), for discriminating non-alarm screams from non-alarm screams (second set), for discriminating alarm screams from non-alarm screams (third set), and for discriminating alarm screams from alarm screams (fourth set). Panels (j) and (m) show the normalized RT and accuracy difference between the 21 combinations of screams. *p < 0.05, based on a non-parametric permutation test. Numerical data underlying the plots (a–m) can be found in S2 Data. 2AFC, 2-alternative forced-choice; 7AFC, 7-alternative forced-choice; ang, anger; fea, fear; neu, neutral; norm diff, normalized difference; pai, pain; ple, pleasure; sad, sadness; sem, standard error of the mean.
Fig 3.
Neural activity and effective functional network for scream call processing.
(a) Behavioral data from the gender task during the functional magnetic resonance imaging (fMRI) experiment (experiment 4); shown are the response times (left) and the accuracy level (right) for performing a gender decision task on the screams perceived during the fMRI experiment (n = 30). (b) Negative parametric modulation of brain activity by the alarm rating of scream calls; while we did not find neural activity for a positive relationship with the alarm ratings of every scream, we found largely extended activity in the auditory cortex and the frontal cortex for negative associations with the alarm level of each scream. (c) Functional definition of the temporal voice area (TVA) in n = 30 human participants. The red dashed line indicates the cortical extension of the TVA; the green dashed line indicates the coverage of the partial volume acquisition; the white dashed line defines the anatomical subregions of the auditory cortex (primary areas Te1.0, Te1.1, Te1.2, and secondary area Te3) and the planum temporale (PTe). (d) Alarm screams compared with neutral screams (n = 30) elicited reduced activity in the bilateral IFC and STC (left mSTG, right mSTS), whereas non-alarm screams elicited increased activity in the bilateral STC (left PPo, right mSTG). (e) Compared with alarm screams, non-alarm screams elicited significantly higher and extended activity in the bilateral IFC and STC. (f) Positive screams (pleasure, joy) showed higher activity in the right pSTS compared with that for neutral screams and in the bilateral amygdala compared with that for negative screams (sadness, pain, fear, anger). (g) Dynamic causal modeling (DCM) revealed the bidirectional model family as the winning family based on the posterior probability (left panel) in both the left (top left) and right hemispheres (bottom left). The right panel shows the posterior probabilities (black) and the log-evidence (blue; “diff F” = F-value minus the minimum F-value across models) for each model from the forward (for), backward (back), and bidirectional (bidir) model families. (h) Bayesian model averaging in the bidirectional model family indicated specific significant modulation of the left (n = 28) and right hemisphere (n = 28) only in the non-alarm (orange) and positive (green) scream conditions (see S3 Table). All activations are thresholded at p = 0.005 voxel level and a minimum cluster size of k = 42, resulting in a corrected threshold of p = 0.05 at the cluster level. Numerical data underlying the plots in (a) can be found in S3 Data and S3 Table. Amy, amygdala; ang, anger; fea, fear; IFC, inferior frontal cortex; L, left; mSTG, middle superior temporal gyrus; mSTS, middle superior temporal sulcus; neu, neutral; pai, pain; ple, pleasure; PPo, planum polare; R, right; pSTS, posterior superior temporal sulcus; sad, sadness; sem, standard error of the mean; STC, superior temporal cortex.