In traditional smart home systems, the effectiveness of voice recognition is often poor due to the influence of noisy environments and user voice expression methods, which cannot accurately understand user instructions, resulting in poor interaction design. This article applied the Transformer model to improve the recognition ability of smart home systems for user instructions, and enhance the accuracy and smoothness of interaction. The voice data between users and smart homes was collected, and the collected voice data was preprocessed, including segmentation of voice segments, noise reduction, and other processing. The Transformer model was established, and the encoded features were mapped to a linear layer in the output space. Text sequence information was output through logistic regression, and users interacted with smart homes through voice control. The experimental results showed that the application of the Transformer model could effectively improve the accuracy of voice recognition, and the interaction design of the Transformer model could effectively enhance the user experience. The average user experience score reached 90.4. The application of the Transformer model can effectively improve the interaction design and user experience in smart home voice control.