Abstract: Oral movement is closely related to people's eating habits. In this paper, an investigation of people's eating habits is conducted through the analysis and identification of oral movement, so as to guide people regulating their eating habits. The speech recognition method is used to analyze and recognize the bone conduction sound produced by oral movement. In this paper, based on hidden Markov model (HMM), a bone conduction sound recognition system is established with the help of HTK toolkit. Before recognizing bone conduction sound, model training is first carried out by windowing on frames and extracting Mel frequency cepstral coefficients (MFCC). The process of model training is to improve model parameters for establishing a template library. In the process of recognition, the model with the highest match to the audio signal to be tested is found in the template library, and the output result of this model is taken as the final recognition result with a recognition accuracy reaching 84.7%. The experimental results show that the method is feasible.