The IIM System for Blizzard Challenge 2019

Husen Shi, Xiao Zhou, Jia Li, Lei Xiao, Zengfu Wang

Abstract: This paper introduces the IIM-USTC speech synthesis system for Blizzard Challenge 2019. The task is to build a speech synthesis system on a 8-hour Chinese male talkshow audio corpus. The submitted system followed our previous one proposed in Blizzard Challenge 2018. A hidden Markov model (HMM)- based unit selection system was built with improvements in back-end acoustic modeling. Two models were built for unit selection, an LSTM-RNN based acoustic model was built and the hidden layer was adopted as context embedding feature, a DNN based unit embedding model was built and the unit vector was adopted as phone unit feature. Evaluation results demonstrated that our system performed at the same level as last year.