I'm developing an audio watermarking algorithm which embeds text (binary form) in DWT-DCT coefficients. The goal of the algorithm is to hide song information (like artist, song, album,...). Everything works fine the text is recovered using another algorithm which extracts the hidden text. However my algorithm doesn't resist attacks, I use Stirmark for Audio v02 which I found here. The quality of the audio is not really affected, in fact I get good values when using the Normalized Cross-Correlation, the problem is when recovering the text which is highly affected.

As I said before the text is binarized so when I perform an attack it is very likely that some bits representing the text are flipped. If this were a binary image and not text, the extraction would be almost imperceptible to human eye but when dealing with text everything screws up.

So my question are: do you there another way to embed text different from binarization? Do you know any other algorithm to hide text on audio that resists common audio attacks? Any other recommendation to hide same data on audio file?

Thank you

  • $\begingroup$ This is a very advanced topic and requires sophisticated methods. The answer is also of significant commercial value, which is why I would not expect to get a lot of useful answers. If you would like to have real help with this you might want to consider hiring a consultant. $\endgroup$ – Jazzmaniac Nov 21 '13 at 17:03
  • 3
    $\begingroup$ If you expect some amount of bit errors in the embedded stream, you might look into encoding the data using some kind of error correction code (ECC). If you can characterize the bit error rate and the amount of overhead that you can tolerate adding to the bit stream, then you can select an appropriate scheme. $\endgroup$ – Jason R Nov 21 '13 at 17:56
  • $\begingroup$ @Jazzmaniac the algorithm it's already coded, the problem I think is more related to best practices, experience in the field, or some kind of ECC like Jason R suggests $\endgroup$ – Jorge Zapata Nov 21 '13 at 18:31
  • $\begingroup$ @JorgeZapatar, I was meaning to suggest that there are more robust methods than what you describe. There's nothing wrong with ECC, but if you have to rely on in for even unhearable signal degradation, then your algorithm might not be the best choice. $\endgroup$ – Jazzmaniac Nov 21 '13 at 18:49
  • $\begingroup$ redundncy rdundancy redundency redunbancy $\endgroup$ – pichenettes Nov 21 '13 at 19:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.