Highly transparent steganography model of speech signals using continuous Genetic algorithm

In this paper, a speech steganography method in the wavelet domain using Indirect Least Significant Bit (ILSB) is proposed which is utilized genetic algorithm. In this method, we first calculate the Discrete Wavelet Transform (DWT) of the sample of host and secret speech signal. After that, wavelet coefficients of the secret speech signal are fitted effectively and efficiently to the host signal wavelet coefficients using a continuous genetic algorithm. In the genetic algorithm, a fitness scaling method is applied to improve the selection operator and adjustment of selection pressure. In addition, a combination of elitism, crowding replacement and incest prevention is used to provide population diversity and to avoid dominant species creation. Due to the quantization error, there are some differences between the secret signal before and after steganography. But these differences have an appropriate Gaussian noise model. We compress these differences using Huffman lossless compression method. The compression rate of such differences approach to the entropy which is derived from Shannon's first theorem. Huffman lossless compression method, cause to small noise. We these compressed differences sent along the stego signal. The experimental results show a transparent stego signal and high embedding capacity. On the other hand, these results illustrate improvement of mean, variance, skewness and kurtosis values in time and frequency domains in comparison with the three LSB, Frequency Masking (FM) and Efficient Wavelet Masking (EWM). In addition, the continuous genetic algorithm increases 10% of the number of suitable locations for embedding in average.