Speech driven video editing via an audio-conditioned diffusion model