LivelySpeaker: Towards Semantic-aware Co-Speech Gesture Generation