Predicting Expressive Speaking Style From Text In End-To-End Speech Synthesis