Write-a-speaker: Text-based Emotional and Rhythmic Talking-head Generation.