High payload audio watermarking: Toward channel characterization of MP3 compression

A new audio watermarking algorithm resilient to MPEG 1 layer III (MP3) with bit rate of 64 kbps is proposed. High embedding capacity is its main characteris- tic. The proposed algorithm is able to use either semi-blind or blind decoding. With the former, the algorithm achieves an embedding capacity of 204 bits per second (bps) and with the latter obtains 155 bps, them both with bit error rate lower than 10^-4. Using semi-blind decoding, benefit of more than 33 bps is obtained in comparison with, in our knowledge, any other previously proposed algorithm for compression at 64 kbps. The algorithm uses dither modulation to embed a coded-watermark in certain frequencies of wavelet domain. Those frequencies were found to be robust against compression. The coded-watermark is generated with the concatenation of low-density parity-check (LDPC) codes and repetition codes. Also, an accurate estimation of the statistics of the channel noise, due to compression, is introduced. Finally the proposed algorithm is improved using erasures at the decoder. This improvement achieves more than 15 bps in com- parison with normal decoding. The highest embedding capacity achieved is 229 bps. All the watermarked audio files obtain SNR higher than 40 dB. The audio quality was also measured with a subjective evaluation in which the watermarked audio obtained a score higher than 4, where the best audio quality is scored with 5.