Audio Watermark Embedding Technique Applying Auditory Stream Segregation: 'G-encoder Mark,' Extractable by Mobile Phone

摘要

We are developing audio watermarking techniques which enable extraction of embedded data by cell phones. For this purpose data must be embedded into frequency ranges where the auditory response is prominent, and therefore data embedding will cause large amounts of auditory noise. Previously we have proposed applying a two-channel stereo play-back feature, in which noise generated by a data-embedded left-channel signal will be reduced by the other right-channel signal. However, this proposal has the practical problem of restricting the location of the extracting terminal. In this paper, we propose to synthesize the noise-reducing right-channel signal with the left-channel signal, which reduces noises completely by making use of the auditory stream segregation phenomenon in human hearing. This new proposal makes the noise-reducing right-channel signal unnecessary and supports monaural playback operation. We also propose a wide-band embedding method causing dual auditory stream segregation, which makes possible data embedding in the entire public phone frequency range, and also stable extraction by 3-G mobile phones. When using this method, the extraction precision becomes greater than that of the previously proposed method, and the quality degradation of the embedded signals is reduced. We present a survey of our newly proposed method, together with a comparison of experimental results comparing the new method and the previously proposed method.