Gesture Generation and Pose-Forecasting for an Avatar-Based Communication System