The invention provides an emotional speech synthesizing method and device. According to an aspect of the invention, the emotional speech synthesizing method comprises the following steps: inputting a text sentence; forecasting a neutral characteristic vector of the text sentence in first characteristic space of a first speaker by utilizing a neutral characteristic model acquired by neutral voice database training of the first speaker; converting the neutral characteristic vector into a regularly neutral characteristic vector in second characteristic space of a second speaker by utilizing a speaker regular model acquired by the neutral voice database and parallel voice database training of the second speaker; converting the regularly neutral characteristic vector into a regular emotional characteristic vector in the first characteristic space by utilizing the speaker regular model; and synthesizing the emotional voice of the first speaker by utilizing the emotional characteristic vector in the first characteristic space.