Research Of Digital Audio Watermarking Based On Spread-spectrum And LPC Prediction Error

The Internet technology makes it easy to distribute digital media to all over the world. Meanwhile, it brings forward new issues on transmission security and intellectual property protection. Copy-right violation and various attacks through the Internet have been more and more serious. Digital watermarking technology is dedicated to labeling the authentic source of a digital production, which is an effective way to resolve the issues of copy-right protection and transmission security.The research of this thesis focuses on checking whether speech signals have been tampered or not, authenticating the integrity of speech signals by means of fragile watermarking, and detecting the spread spectrum watermarking in speech signals. Progresses are summarized as follows:1. LPC (Linear Predictive Coding) technique is introduced to the research and analysis of speech signals. LPC technique is used to estimate the basic parameters of digital speech signals, which aims at obtaining a set of prediction coefficients and prediction error. Since the value of LPCPE (LPC prediction error) can be a good reflection of speech signals' acoustics characteristic, it is regarded as invariable features of speech in this thesis.2. A new method for blind speech authentication is proposed based on LPCPE combined with LSB (Least Significant Bit) steganography. The blind speech authentication method takes LPCPE as invariable features of speech and inserts these features into least significant bits of speech samples. When verifying the speech, we extract the inserted features, and compare them with recalculated LPCPE. Experimental results show that this method can detect not only the tampering in speech but also the position and region of the tampering.3. A new steganalysis method for digital speech watermarking is proposed based on LPCPE. Since value of LPCPE can reflect the acoustics characteristic of speech signals, the steganalysis method concentrates on analyzing the difference of LPCPE before and after embedding watermark information, especially finding out in whichregion the difference is more distinct. Experiments show the method is feasible and effective in digital speech watermarking steganalysis.