Text classification adapted for Multi-domain Text-To-Speech Synthesis