Integrating Self-supervised Speech Model with Pseudo Word-level Targets from Visually-grounded Speech Model.