Albayzin 2012 language recognition evaluation (LRE) is one of the most challenging language recognition evaluation, which is mainly reflected in: (1) the target languages are more confusable with other languages, which might push down the system performance; (2) developing and test data is heterogeneous regarding duration, number of speakers, ambient noise/music, channel conditions, etc. (3) signals may contain noise, background music and any kind of nonhuman sounds. To solve these problem, in Department of Electronic Engineering, Tsinghua University (THUEE) system we develop (1) 47-phoneme English Gaussian mixture model-hidden Markov model (GMM-HMM) decoder with background noise model for voice activity detection (2) noisy and clean model separately and fusing the weighted model (3) linear discriminant analysis-minimal mutual information (LDA-MMI)+Multifocal fusion method to improve the LRE system performance and the system yielded Fact = 0.1513 in empty training closed-set tests.