Speech synthesis by rule based on synthesis units considering prosodic features