An efficient retrieval approach for encrypted speech based on biological hashing and spectral subtraction

Abstract

In this paper, an efficient retrieval approach for encrypted speech based on biological hashing and spectral subtraction is proposed. The proposed approach improves the impact of noise on the robustness and discrimination of speech hashing scheme, as well as improves the retrieval efficiency, accuracy, and security of search digests, and realizes the authentication of the query result. The speech owner firstly secure the original speech file by encrypting it with two-dimensional Arnold mapping and uploading it to the encrypted speech library on cloud. Then, the pre-processed speech signal is subjected to spectral subtraction and noise reduction, as well as the discrete wavelet transform (DWT) is performed to obtain the wavelet low-frequency coefficient and reconstruct the speech signal, calculating the normalized autocorrelation function to obtain the matrix feature vector, and using the Chebychew mapping algorithm to generate the pseudo-random matrix, and generate the pseudo-random Fourier matrix by fast Fourier transform (FFT). Finally, iterate the matrix feature vector and pseudo-random matrix. After the thresholding, the hash sequence is constructed and uploaded to the system hash index table on cloud. When speech’s user retrieval, the Hamming distance algorithm is used for the matching retrieval operation during the search and integrity authentication of the query result. The experimental results show that the proposed approach effectively reduces the noise of speech, with strong robustness and discrimination, and the retrieval efficiency, accuracy and security have been significantly improved.