Exploring Self-Attention Mechanisms for Speech Separation