A closer look at internal representations of end-to-end Text-to-Speech models: How is phonetic and acoustic information encoded?