UniBriVL: Robust Universal Representation and Generation of Audio Driven Diffusion Models