Shallow Diffusion For Fast Speech To Text