Command passwords recognition in complex and strong interference sources

Based on the analysis of the character of Chinese, syllable division and endpoint detection in a high noise environment is achieved using the time-frequency parameter (TFP). Using TFP with RASTA-PLP (Relative Spectral-Perceptual Linear Perception) and SDA (Static, Dynamic and Acceleration), robust speech features are extracted. The differences between classes and the similarities within classes are increased by setting up a multi-value threshold function. In the presence of 80-95 dB noise level recorded in a vessel compartment, the results demonstrate the good performance of the system, and the TOP3 recognition rate of 200 passwords is better than 95%.