Абстрактный

Robust information hiding and extraction algorithms in speech

Bao Yongqiang, Xi Ji, Xu Haiyan

Speech with hidden data will be disturbed and damaged by a variety of interference, such as noises, codec and filters, etc. To improve the robustness, the speech information hiding and extraction algorithmbased on PSO-NN (Particle SwarmOptimizer Neural Network) is proposed. To improve the performance of anti-channel interference, the algorithmadds redundant data into the hidden data and then trains at the decoding end. At the same time, to improve the training efficiency and decoding accuracy, the algorithm firstly uses wavelet decomposition to get high-frequency coefficients of the signal, and then calculates the characteristic of highfrequency coefficients. At last, the algorithm selects 32 optimal features to train the neural network based on the FDR (Fish Discriminant Ratio). Simulation results show that the proposed algorithm improves the robustness of speech information hiding approach against filtering attack, noise attack, sampling attack and compression attack. Though the improvement on tensile attacks is ineffective, it was also better than others neural network algorithm.