A neural network approach to audio data hiding based on perceptual masking model of the human auditory system

摘要

A new system that employs artificial neural networks to identify perceptually masked transmission opportunities within an audio stream is presented. The neural network is trained to automatically extract the perceptual map of the human auditory system. The network is then used at the encoding end to identify opportunities for transmission of inaudible data into the voice stream. At the decoding end, the network is used to monitor the audio channel for presence of masked data. Increased data transmission rates, resistance to compression algorithms and increased processing gains are among the advantages of the proposed solution.