有別於直接將資料做加密（encryption），資料隱藏（data hiding）技術可將資料隱藏在另一個媒體之中再傳送，以提升通訊的安全性。目前針對語音訊號所提出的資料隱藏方法，大多只能抵抗失真較低的壓縮法，因此無法結合失真較大但廣泛應用於通訊系統中的線性預測語音編碼（linear predictive coding）以同時傳送語音及隱藏的資料，所以另一研究方向便是朝已壓縮或正在進行壓縮之語音來隱藏資料。 本論文提出兩種針對MELP及G.729已壓縮語音之資料隱藏法，第一種方法稱作不重要位元取代法（least-significant-bit substitution method），其基本原理是先分析語音經語音壓縮編碼後的音框（frame）每個位元的重要性，再將欲隱藏的資料直接取代那些較不重要的位元，以達到資料隱藏的目的。第二種方法稱作類抖動資料隱藏法（dither-like data hiding method），其原理是運用MELP及G.729本身具備的多級向量量化器（multistage VQ）機制與可去除抖動（subtractive dither）的量化架構相結合，利用可去除抖動在統計特性具有不影響量化誤差和多級向量量化器較後級的碼簿向量（code vector）與雜訊相似的特性，將資料隱藏在多級向量量化器的索引（index）之中。兩種資料隱藏方法皆可做到同時傳送語音及機密資料並能與原來系統相容的目的，並且經由模擬實驗的結果證實，在同樣的資料隱藏位元率下（data embedding rate），類抖動資料隱藏法可達到比不重要位元取代法更低的失真。 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.
