Watermarking of Audio Signals Using Adaptive Subband Filtering and Manchester Signaling

Methods of data hiding are considered, based on subband Alters, enabling the amplification or attenuation of audio signal within each subband. This spectrum modification aims at maximizing the selected cepstrum coefficient. Because the cepstrum coefficient is a scalar product of the signal log-spectrum and a real part of the proper IDFT base function (called here a cepstral kernel), the audio signal is amplified in these subbands, where this real part is positive and it is attenuated in these subbands where the real part is negative. The maximum amplitude modification at a given frequency is limited by the masking threshold at this frequency. This method may be also used with an echo insertion algorithm, thus enhancing the cepstral peak and reducing the bit error rate. Using mirrored cepstral kernels (i.e. a kind of Manchester signaling, consisting in reversing the kernel sign in two consecutive time slots) the influence of high spectral peaks of the audio spectrum is reduced and a robust watermarking system is obtained. Simulation results are given, for hidden data transmission rate of 86 bit/s.