Regions of interest.

Speech processing can be modelled using a dynamic causal model (DCM) with 6 regions. The table lists the central coordinates of these regions in MNI152 space. These coordinates define the centre of the rough anatomical masks (16 mm×16 mm×16 mm) that guided the specification of the exact location and extent of the regions of interest underlying model inversion (see Section ‘Implementation of generative embedding’). For an illustration of these masks, see Figure S1 in the Supplementary Material.