Early blind persons have often been shown to be superior to sighted ones across a wide range of non-visual perceptual abilities, which in turn are often explained by the functionally relevant recruitment of occipital areas. While voice stimuli are known to involve voice-selective areas of the superior temporal sulcus (STS) in sighted persons, it remains unknown if the processing of vocal stimuli involves similar brain regions in blind persons, or whether it benefits from cross-modal processing. To address these questions, we used fMRI to measure cerebral responses to voice and non-voice stimuli in blind (congenital and acquired) and sighted subjects. The global comparison of all sounds vs. silence showed a different pattern of activation between blind (pooled congenital and acquired) and sighted groups, with blind subjects showing stronger activation of occipital areas but weaker activation of temporal areas centered around Heschl's gyrus. In contrast, the specific comparison of vocal vs. non-vocal sounds did not isolate activations in the occipital areas in either of the blind groups. In the congenitally blind group, however, it led to a stronger activation in the left STS, and to a lesser extent in the fusiform cortex, compared to both sighted participants and those with acquired blindness. Moreover, STS activity in congenital blind participants significantly correlated with performance in a voice discrimination task. This increased recruitment of STS areas in the blind for voice processing is in marked contrast with the usual cross-modal recruitment of occipital cortex.