In recent years, robots are beginning to be used in tasks that require interacting with people, which requires that the robot has a lively appearance. This can be done by having the robot perform expressions autonomously. While this might be easy if only non-verbal gestures are used, when both verbal and non verbal messages are emitted we need to ensure that both components complement each other and are properly synchronized. In this work, we present a gesture prediction and synchronization method for social robots. The deep learning-based prediction module generates sequences of labels that represent the type of gestures that should accompany a given utterance. The rule-based synchronization method selects appropriate expressions and finds the right moment to execute each one. The results of the evaluation performed show the viability of the proposed solution, although there are aspects to improve in future works.