In this work, we describe the TL-NTU’s text-to-speech (TTS) system for the Blizzard Challenge 2018. Our efforts are mainly focused on two aspects, which are the front-end text analysis and back-end model training. For the front-end text analysis, we include phonetic, syllable, and word-level linguistic features using lexicon and word-level text analysis for fullcontext feature extraction. For back-end model training, a feed-forward Deep Neural Network (DNN) based phone duration model and a bidirectional long short-term memory (BLSTM) based acoustic model are trained. The performance of our system is assessed by reporting the results of listening tests provided by the challenge organizer.