Campus Bullying Detection Algorithm Based on Audio

With the continuous breakthroughs in various technologies, voice recognition has become a research hotspot. It is a method to detect the phenomenon of bullying in time by detecting whether the campus bullying emotion is contained in the voice. This paper builds a convolutional neural network model to recognize speech emotions. Firstly, pre-process the audio data, then extract the MFCC feature parameters from the pre-processed audio data, and finally design a classification algorithm. This paper selects the CASIA database, which has a total of 300 voice audios, including six emotions: angry, scared, happy, neutral, sad, and surprised. Using fivefold cross-validation to test the performance of the model, the accuracy of the classification algorithm is 68.51%. Finally, the classification algorithm is used to perform emotion recognition on a test sample selected from a campus bullying movie section. This section shows “fear” emotion, and the algorithm judges that the audio shows “fear” emotion. The actual scenes are consistent, indicating that the classification algorithm in this paper has certain stability and practicability.