Keystroke Sound Recognition Based on Gaussian Fitting Segmentation

A key problem in keystroke recognition based on sound signals is that background noise causes fluctuations in the keystroke sound signal, which makes it difficult to accurately portray the true keystroke waveform and thus seriously affects the recognition accuracy. To address this issue, this paper proposes a keystroke data segmentation method based on Gaussian fitting parameters, which effectively improves the accuracy of keystroke sound segmentation and has stronger resistance to uncertain sound signals. On this basis, a feature extraction method combining Mel Frequency Cepstral Coefficient (MFCC) and Cochlear Filter Cepstral Coefficients (CFCC) is proposed; Finally, the BP neural network was optimized using the step coefficient Nonlinear Update of the Beetle Antennae Search algorithm (NUBAS), and single and fused features were input into the NUBAS-BP model for prediction and classification. The experimental results show that the method proposed in this article can recognize 26 types of letter keys on physical keyboards, and achieve a recognition accuracy of 98.85%. This technology has the benefits of strong recognition performance and low cost, which is crucial for boosting information security and the convenience of human interaction.