Multi-task learning of structured output layer bidirectional LSTMS for speech synthesis

Runnan Li; Zhiyong Wu; Xunying Liu; Helen M. Meng; Lianhong Cai

ICASSP 2017accepted0 citations

Multi-task learning of structured output layer bidirectional LSTMS for speech synthesis

Runnan Li, Zhiyong Wu, Xunying Liu, Helen M. Meng, Lianhong Cai

Abstract

Recurrent neural networks (RNNs) and their bidirectional long short term memory (BLSTM) variants are powerful sequence modelling approaches. Their inherently strong ability in capturing long range temporal dependencies allow BLSTM-RNN speech synthesis systems to produce higher quality and smoother speech trajectories than conventional deep neural networks (DNNs). In this paper, we improve the conventional BLSTM-RNN based approach by introducing a multi-task learned structured output layer where spectral parameter targets are conditioned upon pitch parameters prediction. Both objective and subjective experimental results demonstrated the effectiveness of the proposed technique.

BibTeX

@inproceedings{icassp2017_multitasklearnin,
  title = {Multi-task learning of structured output layer bidirectional LSTMS for speech synthesis},
  author = {Runnan Li and Zhiyong Wu and Xunying Liu and Helen M. Meng and Lianhong Cai},
  booktitle = {ICASSP 2017},
  year = {2017}
}