BLSTM-Duration-model

Make duration model for speech synthesis