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?