We investigated the functional characteristics of brain regions implicated in processing of speech melody by presenting words spoken in either neutral or angry prosody during a functional magnetic resonance imaging experiment using a factorial habituation design. Subjects judged either affective prosody or word class for these vocal stimuli, which could be heard for either the first, second, or third time. Voice-sensitive temporal cortices, as well as the amygdala, insula, and mediodorsal thalami, reacted stronger to angry than to neutral prosody. These stimulus-driven effects were not influenced by the task, suggesting that these brain structures are automatically engaged during processing of emotional information in the voice and operate relatively independent of cognitive demands. By contrast, the right middle temporal gyrus and the bilateral orbito-frontal cortices (OFC) responded stronger during emotion than word classification, but were also sensitive to anger expressed by the voices, suggesting that some perceptual aspects of prosody are also encoded within these regions subserving explicit processing of vocal emotion. The bilateral OFC showed a selective modulation by emotion and repetition, with particularly pronounced responses to angry prosody during the first presentation only, indicating a critical role of the OFC in detection of vocal information that is both novel and behaviorally relevant. These results converge with previous findings obtained for angry faces and suggest a general involvement of the OFC for recognition of anger irrespective of the sensory modality. Taken together, our study reveals that different aspects of voice stimuli and perceptual demands modulate distinct areas involved in the processing of emotional prosody.