Data Hiding Techniques for G.729 and MELP Speech Coding

Abstract

Data hiding is the art of hiding secret messages within a multimedia signal. Most data hiding techniques developed today for speech cannot defense the attack of linear predictive coding (LPC), which is widely used in speech communication systems, and that means it is very difficult to transmit the secret messages and speech simultaneously. A different approach is to hide the secret message in the compressed bit stream.
In this thesis, we present two data hiding techniques in compression domain. The first method is called least-significant-bit substitution (LSBS) method. The basic idea of LSBS method is to analyze the significance of each bit of each coded frame, and then substitutes the LSBs with the data to be hidden. The second method is called dither-like data hiding (DDH) method, which utilizes the characteristics of subtractive dithering and the multistage vector quantization (MSVQ) in G.729 and MELP. The secret data is hidden in the index of the MSVQ. The data stream processed by either LSBS method or DDH method is compatible with MELP or G.729 speech coding standard. From the simulation results, both methods can deliver the secret message and the speech signal simultaneously, and the DDH method provides better quality than LSB Substitution Method at the same data embedding rate.