Exploring the Integration of Speech Separation and Recognition with Self-Supervised Learning Representation