A voice synthesis method is provided. The method includes: for each sound model of a plurality of sound models, performing a first matching operation on a user attribute and a sound model attribute of the sound model to obtain a first matching degree for the sound model attribute, and determining a sound model with a sound model attribute having the highest first matching degree as a recommended sound model; for each content of a plurality of contents, performing a second matching operation on a sound model attribute of the recommended sound model and a content attribute of the content to obtain a second matching degree for the content attribute, and determining a content with a content attribute having the highest second matching degree as a recommended content; and performing a voice synthesis on the recommended content by using the recommended sound model, to obtain a synthesized voice file.