Zhuang language speech synthesis optimization method and system based on transfer learning

The invention discloses a Zhuang language speech synthesis optimization method and system based on transfer learning, and relates to the technical field of speech synthesis. The method comprises the following steps: preprocessing an English text to construct an English phoneme data set; the English phoneme data set is adopted to train an English Tacotron2 model, and English Tacotron2 model parameters are obtained; preprocessing the Zhuang language text to obtain a Zhuang language prime data set; the English Tacotron2 model parameters are loaded to the Zhuang language Tacotron2 model, and a Zhuang language prime data set is used to carry out fine tuning on the migrated Zhuang language Tacotron2 model; in the reasoning stage, a Zhuang language text is preprocessed to obtain a Zhuang language element sequence, the Zhuang language element sequence is input into the fine-tuned Zhuang language Tacotron2 model to obtain Mel-frequency spectrum features, and the Mel-frequency spectrum features are synthesized into Zhu