Language identi¯cation (LID) has received increasing interests in the speech signal processing com- munity. With the rapid development of LID technologies, how to fuse the score of multi-systems is growing to be a researching focus. In this paper, we proposed a discrimina- tive framework for LID score fusion. The Heteroscedastic linear discriminate analysis (HLDA) technology is used for dimension reduction and de-correlation, and the Gaussian mixture model (GMM) trained with Maximum mutual in- formation (MMI) criteria is used as classi¯er. Experiments show that the proposed method can improve the perfor- mance signi¯cantly. By score fusion of ¯ve systems, we achieve average cost of 2.10% for 30s trials on the 2007 NIST language recognition evaluation databases.