Weighted error minimization in assigning prosodic structure for synthetic speech door