The invention provides a speaker segmenting clustering method and device. The speaker segmenting clustering method and device are used for solving the problem that when short-time speaker voices are processed in the related technology, as the performance is reduced, a clustering effect of speakers is poor is solved. The method includes the steps that to-be-clustered voices are divided into multiple sub-voice segments; through a weight-probability linear discriminant analysis (WT-PLDA) model, feature information W-vectors of all the sub-voice segments are extracted for I-vectors of all the sub-voice segments obtained based on dividing, and model parameters of the WT-PLDA model at least include a mean vector, a speaker sub-space projection matrix, a speaker-factor hidden variable and a residual factor; through probability linear discriminant analysis (PLDA), the multiple sub-voice segments are circularly clustered multiple times according to the W-vectors of the multiple sub-voice segments after dividing until the multiple sub-voice segments are clustered into two clusters. According to the speaker segmenting clustering method and device, the clustering effect of speakers is improved.