A Multi-Means Speech Encryption Algorithm Based on Improved Zigzag Transformation

The application of deep neural networks has led to a significant increase in the accuracy of automatic speech recognition. When unencrypted speech is transmitted over the Internet, malicious attackers are able to use automatic speech recognition technology to obtain private information from speech, which brings security risks such as personal information leakage and property loss. In this paper, a reversible speech encryption and decryption algorithm is proposed, to realize the encryption and decryption of speech sequences by improving the ZigZag transformation and integrating the Lorenz chaotic map, DNA coding and Arnold map. For encryption, the key of each algorithm is first calculated using the Hash value and external keys. And then, the speech is divided into two parts, they are executed XOR operation separately with the x and y chaotic sequences generated by the Lorenz map. Next, the two parts of the speech are encrypted with DNA coding and Arnold mapping, and a Zigzag rotation is performed respectively before and after splicing the two parts of the speech sequence. Finally, the speech sequence is XOR with the z0 chaotic sequence generated by the Lorenz map, and the final encrypted sequence is obtained. The experiments show that the algorithm proposed in this paper has a large key space, high key sensitivity, and has better performance than the existing algorithms.