Speech comprehension requires that the brain extract semantic meaning from the spectral features represented at the cochlea. To investigate this process, we recorded fMRI signals during natural narrative speech comprehension. We then used voxel-wise modeling and variance partioning to examine the representation of three feature spaces: the sound spectrum, articulation and semantics. Speech comprehension involves hierarchical representations starting in primary auditory areas and moving laterally on the temporal lobe. We also find that both hemispheres represent these features during language comprehension.