Multi-task self-supervised learning for Robust Speech Recognition