Chinese dialect identification may provide an important clue for forensic investigation. This paper has proposed an effective dialect identification model for chinese dialect identification based on improved Long Short-Term Memory(LSTM). The authors extracted Mel frequency cepstral coefficients (MFCC) from speech samples including regional pet phrase collected from six regions in Guizhou province, then added a corresponding regional pet phrase after each voice sample, and then used the sliding window to conduct information overlapping blocking. The singular value of each block was decomposed from horizontal and vertical and high contribution rate feature vectors were retained, and the blocks were combined as the input data of the dialect identification model. Firstly, the LSTM is improved, then a dialect identification model is constructed, and the model is trained and verified by adopting an experiment, so that the width of the sliding window are optimized and the LSTM is compared with RNN. The experimental results show that the model based on improved LSTM is efficient for Chinese dialect identification.