Speech synthesis of Shanghai dialect based on DNN and LSTM-RNN

This paper proposes a speech synthesis method in Shanghai dialect based on deep learning. We firstly build a Shanghai dialect speech corpus for model training. At the same time, we realize a text analyzer for obtaining context-dependent information of Shanghai dialect from Chinese sentence. Finally, we adopt both deep neural networks (DNN)-based method and long short term memory networks-recurrent neural networks (LSTM-RNN) to realize the speech synthesis of Shanghai dialect. Subjective and objective experimental results show that the proposed method can synthesize the Shanghai dialect speech with better voice quality. The speeches synthesized by the LSTM-RNN-based method have better voice quality than that of the DNN-based method.