Open Collections

Security evaluation of transform domain speech scramblers

University of British Columbia

A number of new speech scrambling algorithms have been recently proposed. Few of them have had their security evaluated. Therefore, this work first concentrates on evaluating, on comparable grounds, various analog speech scramblers. Particular attention is placed on the transform domain speech scramblers. Various attacks are applied to the speech scramblers to evaluate their susceptibility to cryptanalysis. To determine the effectiveness of the cryptanalytic attacks, various measures are proposed and evaluated. These consist of objective distance measures and subjective tests. A novel objective distance measure that emphasizes the lower portion of the log power spectral density of the speech is defined and compared to other objective distance measures. A subjective test that not only tests for lexical residual intelligibility but also for syntactic intelligibility is adapted from audiology experiments used to determine how the working memory affects speech understanding. It is shown that real comprehension of speech may be measured with a lexical intelligibility measure alone, in the specific case where the residual intelligibility of the cryptanalyzed speech is high. However, it is believed that overall speech comprehension is better represented by a composite measure made up of both a lexical and a syntactic intelligibility score. Through the subjective tests, and from the objective measures, it was determined that the transform domain speech scramblers are all seriously vulnerable to simple cryptanalytic attacks. In contrast to [9], the discrete cosine transform (DCT) based scramblers were not found to be more secure than the discrete Fourier transform (DFT) based scramblers. Finally, a noisy modulo addition approach to transform domain speech scrambling is proposed. The scrambling scheme consists of adding pseudo-random noise modulo a certain constant to the quantized spectrum of the speech signal. This method scrambles speech in such a way that it resembles white gaussian noise, and thus traditional cryptanalysis techniques cannot be used to attack it. Its performances under a variety of conditions are discussed.

Thesis/Dissertation

application/pdf

2009-02-14

For non-commercial purposes only, such as research, private study and education. Additional conditions apply, see Terms of Use https://open.library.ubc.ca/terms_of_use.

http://hdl.handle.net/2429/4581

Electrical and Computer Engineering

University of British Columbia

UBCV

DSpace