A method and system for non-automatic regression speech synthesis based on deep neural networks is presented. A deep neural network-based non-automatic regression speech synthesis system according to an embodiment includes a sentence data analysis unit configured to analyze sentence data and output refined sentence data; A speech feature vector sequence synthesis unit that generates a template feature input and generates speech feature vector sequences by adding the refined sentence data to the generated template using an attention mechanism; And a speech reconstruction unit that converts the speech feature vector sequence into speech data.
展开▼