During the last trimester of human gestation, neurons reach their final destination and establish long- and short-distance connections. Due to the difficulties obtaining functional data at this age, the characteristics of the functional architecture at the onset of sensory thalamocortical connectivity in humans remain largely unknown. In particular, it is unknown to what extent responses evoked by an external stimulus are general or already sensitive to certain stimuli. In the present study, we recorded high-density event-related potentials (ERPs) in 19 neonates, tested ten weeks before term (28-32 weeks gestational age (wGA), that is, at an average age of 30 wGA) by means of a syllable discrimination task (i.e., a phonetic change: ba vs. ga; and a voice change: male vs. female voice). We first observed that the syllables elicited 4 peaks with distinct topographies implying a progression of the sensory input along a processing hierarchy; second, repetition induced a decrease in the amplitude (repetition suppression) of these peaks, but their latencies and topographies remained stable; and third, a change of stimulus generated mismatch responses, which were more precisely time-locked to event onset in the case of a phonetic change than in the case of a voice change. A hierarchical and parallel functional architecture is therefore able to process environmental sounds in a timely precise fashion, well before term birth. This elaborate functional architecture at the onset of extrinsic neural activity suggests that specialized areas weakly dependent on the environment are present in the perisylvian region as part of the genetic endowment of the human species.