Close neighbor principle based speaker recognition method for synthesizing emotional model

The invention discloses a close neighbor principle based speaker recognition method for synthesizing an emotional model, which comprises the following steps of: (1) training reference voice and user neutral voice models; (2) extracting an neutral-emotional Gaussian component mapping set of a GMM (Gaussian Mixture Model) reference model; (3) extracting a plurality of neutral reference Gaussian component mapped emotional reference Gaussian components close to user neutral training Gaussian components; (4) synthesizing user emotional training Gaussian components, and then obtaining user emotional training models; (5) synthesizing GMM training models of all users; (6) inputting test voice for recognition. The speaker recognition method adopts a KL (kullback-leibler) divergence based close neighbor principle to extract a plurality of reference voice similar to user neutral training voice from a voice database, and user emotional training voice is synthesized by the emotional reference voice of the reference voice, accordingly, the performance of a speaker recognition system under the mismatching condition of the training voice and the test voice is improved, and the robustness of the speaker recognition system is improved.