A novel data-hiding technique for digital audio is presented in this paper. The proposed audio data-hiding scheme exploits the insensitivity of the human auditory system (HAS) to the phase-distortion, in certain frequency range (f > 3 kHz), for data-hiding. Higher-order allpass filters are used to introduce inaudible phase-distortion in the host audio for information embedding. To improve the embedding capacity of the proposed scheme, the host audio signal is decomposed into subband signals, some of which are selected for data embedding based on the HAS response to phase-distortion to ensure fidelity and robustness of the embedded information. A set of allpass filters, with distinct pole-zero pairs, is used as a codebook for information encoding. The data detection is based on allpass filter parameter, i.e. pole-zero pair, estimation. The detector uses estimated power spectrum of the received audio (data-embedded audio) for the filter parameter estimation. The parametric spectral estimation schemes, such as, autoregressive (AR) and moving-average are used for spectrum estimation of the received audio signal. The detector uses the finite-length impulse response (FIR) approximation of the allpass impulse response allpass filter parameter estimation, for information detection. Experimental results have shown that of the proposed data-hiding scheme can withstand against standard data manipulation attacks. Moreover, the proposed scheme can embed 5-8 times more data then the existing audio data-hiding schemes while providing the same perceptual and robustness embedding performance.

Audio clips used for data hiding based on the proposed APF based data hiding are listed in Table 1

TABLE 1

SELECTED AUDIO CLIPS

FIDELITY PERFORMANCE