The invention relates to the speed emotion recognition technology, in particular to an emotion recognition method based on speech fuzzy clustering. The method includes the steps of carrying out preprocessing on input speech signals, extracting feature information of the processed speech signals, grouping various emotions and selecting corresponding feature information according to the types of the grouped emotions, classifying the feature information combined and selected according to each group of emotion combinations, and carrying out speech emotion recognition on output results after each group of the emotions are combined and classified. The method has the advantages that different features are selected according to different emotions, and compared with an FCM method that all emotions shares the same feature in a traditional mode, an improved self-adaptive fuzzy K mean clustering method is good in recognition effect, high in recognition rate and good in effect. The method is particularly suitable for speech intelligent emotion recognition.