Hiding inaudible data in an audio signal (watermarking), which can be useful for transmitting metadata and inserting copyright identification, is based on taking advantage of the complexity of human perception. In the proposed approach, data is embedded using the polarity of reverberation (echoes) that is added to the high frequencies. A detector uses the sum and difference of the high frequencies in the two channels to extract the data. Average data rates of 22 bps were achieved. The approach was robust with additive noise at 30 dB SNR. In addition MP3, AAC encoding, sample-rate conversion, pitch shifting, and frame misalignment did not destroy the information.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.