Statistics for Improving audio-driven visual dubbing solutions using self-supervised generative adversarial networks