Speech watermarking for mobile communications

Abstract

Watermarking techniques have been widely employed in digital media, such as documents, pictures and speech. In this digital age, mobile phones these days include more functions and have become a commonly used device for communication. Industries, therefore, expand their servicing mode to include phones as a servicing method. Such as banks, telecommunications companies and hotels allow clients to apply and change their services over the phone. Banks, which require authentication from clients, can increase their security to client by applying speech watermarking. The purpose of this project is to seek a methodology for speech watermarking that is more robust in noisy environment. Existing watermarking techniques are studied. The paper by Coumou and Sharma was studied and implemented. The algorithm was tested in additive noise environment. Studies were also done on Quantization Index Modulation, as this was the technique employed in the paper. After which, methodologies were proposed and implemented. The proposed methodologies implemented aims to achieve higher resistance to unauthorized detection. They are also tested in additive noise environment for robustness. All implementation are done in PRAAT. PRAAT is a free program that allows manipulation of pitches and intensity of speech file. The proposed methodologies includes modifying pitches with step sizes based on intensity, using 4 step sizes and to allow user to enter the step size is based on mean of intensity. The results of the proposed methodologies were not as ideal as expected. With each method having its pros and cons, they can be further improved or part of the methodologies could be implemented in other techniques. Though the proposed methodologies were not as ideal as expected, further work and studies can be done on these techniques. Instead of modifying the pitches itself, the intensity of the speech could be modified. The method that uses intensity to determine the step size could have a different step size selection so as to improve the randomness of the step size selected. To conclude, more studies could be done to speech watermarking so as to make it more robust and more resistance to unauthorized detection.