Research on Intention Recognition Methods based on Deep Learning

Abstract

In order to improve the accuracy of intelligent speech interaction robots, the author proposes a deep learning based intention recognition method for research. By introducing the GloveBibGRU-Self attention classification prediction model, an intention recognition function module is constructed, and the ROS distributed architecture is adopted to integrate the system functional modules, achieving intelligent voice interaction between humans and machines. The simulation results show that the speech intention recognition using the proposed method has higher accuracy. Compared with the intention recognition methods based on DCNN model, CNN-LSTM model, and GRU Self attention model constructed unidirectionally, the recognition accuracy is higher than 8. 02%, 4. 06%, and 2. 13%, respectively, and has better recognition effect, In terms of feature extraction, the training time of BiGRU is shortened by four times compared to traditional extraction methods based on BiLSTM models, resulting in higher training efficiency. According to the experimental findings, the speech interaction system developed utilizing the suggested intention recognition method maintains a high level of accuracy and efficiency in understanding user English speech commands. With an average accuracy rate of 89.72% and recognition times consistently below 0.35 seconds, it is evident that the proposed method is applicable for real-world speech interactions. The intent recognition method based on Glove2BiGRU-Self attention can be applied to English speech interaction in intelligent speech robots.