Pitch Delay Based Adaptive Steganography for AMR Speech Stream

Abstract

Most existing speech steganography breaks the continuity of adjacent pitch delay, which obviously degrades their statistically undetectability. This paper presents a novel steganographic scheme for low bit-rate speech stream against pitch delay steganalysis. Three measures are adopted to enhance steganographic security. First, the short-term stability of pitch delay and the statistical distribution of adjacent subframe are considered for designing a distortion function. Second, syndrome-trellis codes (STCs) is utilized to minimize the overall embedding impact based on the defined distortion function. Third, the suboptimal pitch delay is searched to maintain speech quality. Experimental results demonstrate that our scheme achieves higher level of security, especially in the case of low embedding rate. When the relative embedding rate is 0.2 for 10.2 kbit/s AMR stream, the test error rate of our method rises by 12.44% compared with the existing algorithm.