All products featured on WIRED are independently selected by our editors. However, we may receive compensation from retailers and/or from purchases of products through these links.
PITTSBURGH -- Microsoft has developed a prototype system that limits unauthorized playback of music by embedding a watermark that remains permanently attached to audio files.
During a security workshop on Friday, a Microsoft Research scientist demonstrated how the hidden copyright fingerprint is so securely affixed to the audio that it remains intact even if a jazz song is played aloud on speakers in a noisy room and then re-recorded.
Such technology could be included as a default playback mechanism in future versions of the Windows operating system. If the music industry begins to include watermarks in its song files, Windows would refuse to play copyrighted music released after a certain date that was obtained illegally.
Microsoft Research has also invented a separate watermarking system that relies on graph theory -- a concept borrowed from computer science -- to hide watermarks in software.
The paper presented at the Fourth Information Hiding Workshop, principally authored by Microsoft's Ramarathnam Venkatesan, says analyzing program flow through a graph and then subtly altering it to hide a secret watermark "seems to be a viable scheme for large applications."
Probably the most likely application is using the watermark to encode licensing information, such as date of expiration or Ethernet addresses of computers permitted to run the program. When a user double-clicks on a program, Windows would read in the license information and could prevent the program from executing.
A so-called graph is a term that computer scientists use to describe a set of vertices and edges. The paper, entitled "A Graph Theoretic Approach to Software Watermarking," uses a graph representing the flow of a computer program to encode a watermark by adding extra edges.
Pirates hoping to remove the watermark can, of course, try to analyze the application through existing tools like Ursoft's W32Dasm and Optimix, some of which already calculate flow graphs.
The watermarking scheme concedes that a pirate who analyzes every line of a mammoth program like Microsoft Word by hand could remove the hidden information. "Such an adversary can 'undo' any watermark, and our goal is to ensure that this is the only possible model of an effective adversary," says the paper, co-authored by Vijay Vazirani at Georgia Tech and Saurabh Sinha at the University of Washington.
To encode audio watermarks, a pair of Microsoft Research scientists, Darko Kirovski and Henrique Malvar, took a far different approach.
Instead of hiding information in graphs, their approach relies on "spread spectrum" techniques to conceal watermarks in music.
Kirovski, who demonstrated the technique on Friday, said Microsoft hoped to ally itself with record labels and spur e-commerce in music. He said of current online practice, where MP3 files are frequently traded: "All this is super convenient for the user, but is super inconvenient for the record labels, which are losing $5 billion to piracy every year."
"Watermarking may help encourage e-commerce in multimedia," Kirovski said.
He recorded a jazz clip while including his own voice at the same time, and showed that the watermark remained intact.
Here's how the system works: A sound file's creator runs a Windows program that uses spread spectrum techniques to randomly spread the watermark throughout the audio clip by constantly varying the frequency.
The result? A music file that sounds the same as the original, but has the watermark near-irretrievably encoded in it.
After embedding the watermark, a music creator would then digitally sign the audio clip, which means that any tampering with the file would invalidate the digital signature. A future Windows media player might read both the watermark and the signature before deciding whether to play the song.
That would mean future watermarked music released on CDs would be rejected by this hypothetical Windows media player if ripped to an MP3 file, since the digital signature would be absent. Other possible applications include portable audio players that play only music without watermarks, or music with watermarks with the appropriate license.
Kirovski said the watermark is embedded between 2 kHz and 7 kHz and showed that it could not be removed through standard techniques like reverb, echo, hiss reduction and noise reduction. He said even a desynchronization attack -- that varied pitch by up to 5 percent and time by up to 10 percent, and included cutting and pasting chunks up to 100 milliseconds -- was unsuccessful.
The paper he co-authored with Malvar concludes: "We built a data hiding system able to detect covert messages in audio with very high reliability, even in cases where soundtracks are modified by a composition of attacks that degrade the original characteristics of the recording well above the hearing threshold."
But, Kirovski said, it's still a prototype: "We're still not confident enough to put this in a product," he said.
