Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from U-Z

Watermarking, Audio - Challenges, attacks and benchmarking, Applications

data digital example methods

Definition: Digital audio watermarking is a technology to embed and retrieve information into and from digital audio data.

Audio watermarking uses common watermarking methods explained in detail in the article on Digital Watermarking. In this article, the differences to watermarking algorithms for other media types like images and video are discussed.

Raw audio data is commonly stored as PCM (Pulse code modulation) samples. Common sample sizes vary from 8 to 24 bits, sampling rates range form 8 kHz to 96 kHz. Most audio watermarking algorithms work on raw audio data. But there are others which aim at lossy compression audio formats which are commonly applied today for storage or transfer. Mp3 is the best know example for this. Some of these algorithms embed the information in some format specific information, like for example the scale factors of mp3 files. Other algorithms change the compressed files in such a way that the watermark can also be retrieved from the raw data. This is called bit stream embedding.

Common methods like LSB, pattern, statistical methods, patchwork methods, correlation of embedded noise applied in image watermarking are also known for audio watermarking. In audio watermarking, individual bits are often embedded in time frames like shown in Figure 1.

But there are also approaches using specific features of audio data, like echo hiding or phase coding. Echo hiding utilizes the fact that human perception cannot perceive sounds of small energy shortly after loud sounds. An echo hiding watermarking algorithm repeats small amounts of the cover when loud sounds occur with reduced energy. The delay between sound and copy can be used to transfer information. Phase coding is based on the fact that human perception for audio phases is weak, so changes in the phase can be used for transferring information.

Audio watermarking usually addresses watermarking of sound data. But there are also solutions for watermarking musical scores and MIDI data. Musical score watermarking is more similar to watermarking of simples images or drawings due to the low complexity of the cover data. MIDI watermarking also needs to deal with a small cover files with only few possibilities of data modification. LSB watermarking methods of attack velocity codes and small timing modifications have been introduced as suitable watermarking strategies for MIDI data.

Challenges, attacks and benchmarking

At audio watermarking, usually two types of attacks against the watermarking robustness are important: Lossy compression and analogue transfer. Most audio files are distributed in a lossy compression format as for example mp3, wma or ogg. The watermark embedded in the audio data should survive the different compression formats and bit rates as long as an acceptable audio quality is provided. Audio often leaves the digital domain when it is consumed or used, for example in radio networks, live recordings or tape copies. Some applications require the audio watermark to stay in the audio signal and be still retrieval when transferred back into the digital domain. Current state of the art watermarking algorithms provide a good robustness against both attacks.

The growing number of attacks against watermarking systems has shown the importance of efficient and reliable benchmarking to improve the quality of existing watermarking methods. General aspects on attacks and benchmarking can be found in the article on Digital Watermarking. A wide range of image watermarking evaluation approaches and benchmarking suites have been described in the literature by neglecting audio watermarking.

With StirMark Benchmark for audio, a well-defined benchmark for audio watermarking robustness and security has been introduced. The benchmark contains of a set of single geometric attacks in time and frequency domain. They simulate different signal processing effects by adding or removing signals or applying filtering, common in several watermarking applications. The set of attacks allows determining robustness or fragility of a watermark embedded into the audio signal to specific single manipulations or to its arbitrary combination.


While all watermarking applications known from image and video watermarking can also be applied to audio watermarking, the protection of copyright always had an important role in audio watermarking. The best-known example for this is the Secure digital music Initiative (SDMI)) which was a group of companies planning to develop a framework for secure digital music distribution. They tried to install a framework for protected playback, storing, and distribution of digital music. SDMI participants include music content, consumer electronics, information technology, and wireless telecommunication companies. Major music labels where as well included as for example the developers of the Napster software, but complains had been raised that only a small group of participants were involved in key-problem identification and management.

Watermarking, Video - Benchmarking of Video Watermarks [next] [back] Watching the Screen - Fan Magazines, Trade Papers, The MPPDA

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or