Investigation of Text-to-Speech-based Synthetic Parallel Data for Sequence-to-Sequence Non-Parallel Voice Conversion