The invention relates to a voice synthesis device, comprising a voice construction module, a receiving module, a tone processing module, a model correction module and a synthesis module, wherein the tone processing module is used for generating tone information for affecting synthetic voice according to state information indicating the emotion state in the received voice text to be synthesized; and finally, synthetic voice data carrying tone is synthesized via the synthesis module, so that the synthesized voice is highly natural and then the user experience is improved.