Based punishment speaker-independent speech emotion recognition speaker

The present invention discloses a penalty based on speaker independent speaker speech emotion recognition of speech signal samples sequentially pretreated original speech emotion feature extraction, dimensionality reduction, classifier classification decision. In which Jane stage dimensionality, use based on the speaker's punishment graph embedding learning method, using the speaker label information, respectively, for the same class category of sentiment but of different speakers, as well as belonging to the same speaker but belong to different emotional categories speech signal sample pairs, using the existing optimization operation on the basis of the theory FIG embedding theory, be combined. Compared with conventional methods, the method of the present invention is speaker independent speech emotion recognition, can effectively improve the recognition performance of the system.