Due to the difficulties of speech signal processing, there is still a considerable gap between the ability of machines to correctly process and that of human beings. In order to overcome the defects of isolated learning and noise sensitivity of SOM, this paper proposes a new time self-organization model (TSOM) from the perspective of deep learning. On the basis of self-organizing mapping network, time enhancement mechanism is introduced to improve the system performance. This method makes up for the fixed spatial topology of the original self-organizing mapping network and the neglect of the time factor, which is crucial to the voice signal. At the same time, this paper makes full use of computer-aided technology and rich network resources to provide a comprehensive and systematic English pronunciation learning database and establish learners' pronunciation files. Once learners understand and master the operation of voice analysis software, they can conduct self-assessment and judgment to find out their blind spots and weaknesses in voice acquisition.