Dynamic speech MRI is a powerful tool to characterize complex speech articulations. Current accelerated 2D dynamic speech MRI schemes can achieve high time resolutions of the order of (10-20 ms) while sequentially acquiring multiple 2D slices. However, the complex articulatory motion is difficult to interpret jointly across the slices due to mis-alignment of motion patterns. Here, we apply a novel generative manifold model which reconstructs and generates a time aligned multi-slice 2D speech dataset at 18 ms/frame from under-sampled k-space v/s time data sequentially acquired from multiple 2D slices. We evaluate this scheme on two speakers producing repeated speech tasks.