Cerebral processing of emotional prosody—influence of acoustic parameters and arousal

Abstract

The human brain has a preference for processing of emotionally salient stimuli. In the auditory modality, emotional prosody can induce such involuntary biasing of processing resources. To investigate the neural correlates underlying automatic processing of emotional information in the voice, words spoken in neutral, happy, erotic, angry, and fearful prosody were presented in a passive-listening functional magnetic resonance imaging (fMRI) experiment. Hemodynamic responses in right mid superior temporal gyrus (STG) were significantly stronger for all emotional than for neutral intonations. To disentangle the contribution of basic acoustic features and emotional arousal to this activation, the relation between event-related responses and these parameters was evaluated by means of regression analyses. A significant linear dependency between hemodynamic responses of right mid STG and mean intensity, mean fundamental frequency, variability of fundamental frequency, duration, and arousal of the stimuli was observed. While none of the acoustic parameters alone explained the stronger responses of right mid STG to emotional relative to neutral prosody, this stronger responsiveness was abolished both by correcting for arousal or the conjoint effect of the acoustic parameters. In conclusion, our results demonstrate that right mid STG is sensitive to various emotions conveyed by prosody, an effect which is driven by a combination of acoustic features that express the emotional arousal in the speaker’s voice.

Introduction

In a multifaceted environment, our sensory systems are confronted with an abundance of information which contrasts with the limited processing capacity of the human brain. To cope with these limitations, stimuli of potential behavioral relevance have to be separated automatically from irrelevant ones to enable filtering of vital information. This separation can either occur by voluntary attention to certain stimulus features or in an involuntary stimulus-driven manner (Desimone and Duncan, 1995). Such automatic processing has been observed in response to novel stimuli (Näätänen, 1990) or emotionally salient stimuli (Vuilleumier and Schwartz, 2001, Vuilleumier, 2005).

In the auditory domain, emotionally salient information can be expressed via modulation of speech melody (emotional prosody). Findings obtained from auditory evoked potentials indicate differential processing of emotionally and neutrally spoken vowels around 400 ms after stimulus onset (Alter et al., 2003). In previous functional magnetic resonance imaging (fMRI) experiments, it has been demonstrated (Grandjean et al., 2005, Ethofer et al., 2006a, Beaucousin et al., 2007) that voice-sensitive regions (Belin et al., 2000) in the associative auditory cortex in the middle part of the superior temporal gyrus (mid STG) adjacent to superior temporal sulcus respond stronger to happy and angry than to neutral intonations. This response pattern was found independently of whether the subjects were instructed to attend to emotional prosody (Ethofer et al., 2006a, Beaucousin et al., 2007) or some other feature of the presented stimuli, such as emotional semantics (Ethofer et al., 2006a) or speakers’ gender (Grandjean et al., 2005). These findings suggest that activity in mid STG regions is mainly driven by bottom-up mechanisms which rely on features of the presented stimuli and is relatively unaffected by top-down processes that focus the attention of the subjects on certain stimulus characteristics.

So far, it is unclear which stimulus-bound features mediate these bottom-up mechanisms and whether stronger responses to emotional as compared to neutral prosody can be explained by differences in one single acoustic parameter. To minimize such biases, previous studies investigating the neural correlates underlying perception of emotional prosody employed stimuli which were matched for basic parameters, such as acoustic energy (Grandjean et al., 2005) or maximum peak intensity (Ethofer et al., 2006a). Obviously, such approaches are limited by the fact that emotional information in the voice is transmitted via certain acoustic features (Banse and Scherer, 1996) and matching emotional stimuli for all possible acoustic parameters to neutral stimuli would presumably remove the emotional information conveyed by prosody. Therefore, an alternative approach was employed in the present study: first, the impact of single acoustic parameters or the conjoint effect of a set of acoustic parameters was evaluated in simple or multiple regression analyses, respectively. After removing all the variance correlating with the parameters in question, we tested whether the respective regression residuals still show the effect of stronger responsiveness to emotional than neutral prosody.

Here, we used event-related fMRI to investigate the neuronal correlates underlying automatic processing of a broad variety of prosodic categories expressing highly arousing emotional information. To this end, we presented words spoken in neutral, happy, erotic, angry, or fearful intonation in a passive-listening paradigm. Based on results of previous neuroimaging experiments (Grandjean et al., 2005, Ethofer et al., 2006a, Beaucousin et al., 2007), we hypothesized that voice-sensitive regions in mid STG show stronger blood oxygen level dependent (BOLD) responses to emotional intonations expressing high emotional arousal than to low arousing neutral intonations. To evaluate whether stronger responsiveness of this region can be explained by the perceived emotional arousal or by basic acoustic parameters, such as mean intensity, variability (standard deviation) of intensity, mean fundamental frequency, variability of fundamental frequency, or stimulus duration, separate simple regression analyses were carried out and the resulting regression residuals of emotional and neutral trials were statistically compared. Furthermore, a multiple regression analysis, including all five acoustic parameters investigated here, was conducted and regression residuals of emotional and neutral trials obtained in this analysis were compared to evaluate the conjoint effect of these parameters on event-related responses of voice-sensitive regions in mid STG.