Acoustic path is one of the important components of acoustic system. The effective identification of acoustic path can effectively improve the quality of sound. The development of traditional acoustic identification methods has been relatively mature, including Wiener filter, which is widely used in known linear systems, adaptive filter, least mean square filter, and so on. However, with the complexity of the system and the improvement of the accuracy of the model, the traditional identification methods are difficult to apply to this scenario. With the development of deep learning technology, network frameworks with different functions are constantly emerging and upgrading. Among them, long short term memory (LSTM) network has good nonlinear mapping ability of neural network, which can solve the identification problem of unknown uncertain nonlinear systems. Therefore, the acoustic system identification method based on deep learning framework is studied in this paper.