Variance-Preserving-Based Interpolation Diffusion Models for Speech Enhancement