At present, when smart homes are widely used, the security of voice interaction is very important, and the secure transmission of voice data is the key to ensuring user privacy and system security. This study for intelligent home voice interaction engine audio data transmission security problems, put forward based on deep learning and pattern recognition security strategy, through the improved CNN model learning audio features and adopts the new encryption algorithm coding after a large number of real environment test, can ensure the accuracy of the interaction at the same time improve transmission security. Research also discusses the different noise backgrounds of voice interaction engine performance, comparative analysis of the common security protocol efficiency, data show that the strategy in the environment of dynamic noise robustness is better than the traditional method, verified its innovative and practical, the future is expected to be widely used in smart home products to ensure the user information security.