Steganography In Digital Speech And High Fidelity Audio Signals

With rapid development of the information technology and computer science, information security has become a crucial issue that is closely related to social stability and national security. As a growing area of information security research, information hiding, including digital watermarking and steganography, has received enormous attention from the academia, the industry, and various organizations. Watermarking aims to protect intellectual property rights of multimedia contents, while steganography is to send secret message under the cover of carrier signals. By inserting secret data into digital media such as image, audio and video, one can send a secret message imperceptibly through public channels, especially the Internet. As an important component of the digital media, digital speech and high fidelity digital audio can be used as covers for such hidden communications. This dissertation studies steganographic techniques that use different forms of digital audio signals as covers, including linear PCM speech, compressed speech and compressed audio coded bit streams, and proposes three new emedding schemes. The research findings reported in this dissertation are summarized as follows.1. A novel steganographic method is proposed based on a dynamic significant bit (DSB) modification scheme. With DSB, the embedded data are robust against A law andμlaw compression coding, and can resist conversion between these two laws so that the stego-speech can survive transmission through the Internet and PSTN. Experimental results show that, when the embedding capacity is below 20%, difference between the stego-speech and the carrier is imperceptible under subjective testing, histogram comparison, and spectrogram observation.2. An embedding technique is developed to hide secret data in GSM full rate transcoding bit-streams. In this method, the second bits of Regular Pulse Excitation (RPE) samples are replaced with the secret bits and the third bits of RPE are jointly altered sometimes in order to reduce the distortion induced in the data embedding. When the embedding rate is at 550bits/s or below, perceptual quality is not affected. Neither subjective testing nor spectrogram observation can distinguish stego-speech from the host. The MOS metric approaches 3.3 in the objective assessment of PESQ, indicating that the stego-speech can meet the communications quality. Channel coding of the GSM full rate services has been used so that the hidden data are protected in wireless transmission.3. Secret information is embedded directly into MP3 data streams based on a detailed analysis of the MP3 data structure. The proposed scheme can hide a fairly large amount of data. For example, compared with the popular MP3stego, the embedding capacity is four times higher. The embedded information is also highly imperceptible. The stego-audio can pass both listening test and objective assessment. The Objective Difference Grade (ODG) of stego-audio signals is only a few percent less than the host-audio signal.As a basis in the investigation, various subjective and objective methods for assessment of audio quality are studied. PESQ and PEAQ have been chosen respectively in assessing the quality of stego-speech and stego-audio in the experiments. Objective assessment methods can be used to a large extent in the research of audio steganography instead of the expensive and time-consuming subjective assessment to avoid excessive cost and save time.