Dither-like data hiding in multistage vector quantization of MELP and G.729 speech coding

We present a speech data hiding technique that utilizes the characteristics of multistage vector quantization (MSVQ) and subtractive dithering to maintain high speech reconstruction quality. The last stage of MSVQ is used to store the data to be embedded. Similar to subtractive dithering, the noise-like hidden data is subtracted from the signal at the first stage of the encoder and added back at the MSVQ decoder. As a result, the degradation caused by hiding secret data is significantly reduced compared to the traditional simple substitution method.