Speech Recognition Algorithm for Smart Home Scenarios Based on Multimodal Information Fusion

With the rapid development of smart home technology, how to improve the accuracy and efficiency of speech recognition in smart home environments has become an urgent problem that needs to be solved. This article first uses convolutional neural networks to process the obtained visual information and analyze the speaker's facial movements and mouth movements; at the same time, long short-term memory networks are used to process audio signals in order to capture time series features of speech. Next, the information from these two modalities is combined in a fusion layer and weighted to optimize the information weights between modalities. At distances of 2 meters and 3 meters, the accuracy rates were 90.3% and 85.0%, respectively, indicating that the system still maintains high recognition performance at these distances. Through this study, we demonstrate the practical application value and broad potential of multimodal information fusion technology in smart home speech recognition scenarios, providing feasible technical paths and theoretical support for the optimization of future smart home systems.