Audio Watermarking Using Interpolation
Ryan Coppard
A new way to embed a secret signature onto your audio signals for copyright protection.
This project takes advantage of the nature and shape of an audio signal to embed maximally robust watermarks in the time-domain. Typical time-domain methods are vulnerable to attacks which can remove the watermark. My Interpolated Samples Method embeds information strategically so that the shape of the signal is not altered enough to cause an audible difference. Among many advantages, this allows the embedded information to be larger and more robust against attacks which might otherwise erase a typical time-domain watermark.
Page Views:
Audio watermarking is a bit different from image watermarking. While image watermarks might be plainly seen, an audio watermark should never be heard. The two main methods of audio watermarking involve adding or changing the frequency content of the signal or altering the signals specific sample values. This new method, the Interpolated Samples Method (ISM), falls into the latter category.
The ISM works by using a curve fitting method to restore the shape of the signal around a number of random sample points. For each of these points, it then travels randomly along the curve a slight distance away from the original sample and selects that amplitude to be the new amplitude of that point.
By altering the value of a point this way, we ensure that the audio signal retains a more natural shape which is less likely to become audibly distorted. Because of this we can make quite large alterations to an audio signal when watermarking, without changing the quality of the audio.
It’s these comparatively large changes to sample points that set the ISM apart from other time-domain watermarking methods, allowing it to be more robust against attacks and alterations, while still difficult to detect due to its random nature.
