Lightweight End-to-end Text-to-speech Synthesis for low resource on-device applications