Natural speech translation system

The invention discloses a natural speech translation system, which comprises the steps that an interactive interface stores speech data submitted by a user to a data storage service, and the speech data at least comprises speech to be translated; the modular algorithm service pulls the voice data from the data storage service to perform human voice detection, and performs voice recognition, parangulation recognition and text translation on the to-be-translated voice after judging that the to-be-translated voice meets a preset requirement according to a human voice detection result to obtain a translated text with parangulation information; calling a text language model to extract semantic features of the translated text with the sublanguage information, calling a multi-modal voice model to process the to-be-translated voice and the semantic features to generate translated voice data with the sublanguage information and the multi-modal voice information, and submitting the translated voice data to an interacti