
MP3 Watermarking
Wednesday, May 31st, 2006

After the issue lately has come up a couple of times in several forums, some words on the principle of watermarking may be helpful. Digital watermarking describes the process of embedding additional data into some digital document, like author or copyright information, without decreasing the document’s value. Taking it one step further, you wouldn’t want anybody besides you to see that embedded data, which is achieved by using a special watermarking technique called steganography, meaning hiding information within a digital document. I’ll use the term watermarking synonymically for both the general and the hidden approach thereafter.
MP3
Speaking of audio documents, aka tracks, the document type in question is mp3, as this is the common exchange format. MP3, or more specifically MPEG-1 Layer 3, as it is correcty called, is a lossy compression technique and file format, based on two internal compression steps. The first one takes into account certain properties of the human auditory system (HAS) to discard a certain amount of the original data.
When we hear a loud signal of a certain frequency, a second signal of similar frequency but lower amplitude taking place at the same time is masked, meaning we hardly perceive it. In a similar way, there also are timing masking effects. The MP3 encoding algorithm uses these properties of the HAS to reduce the stored data, while trying to keep the same perceptable audio quality. The user can ultimately determine this degree of perceptable quality by choosing a certain bitrate, which expresses the number of bytes the encoding process can use for storing a certain time interval of the source audio stream. By this, the audio data is packed into little pieces of data, called frames, which as a sequence make up the MP3 data stream. This also allows for streaming audio data on the internet as in IP radio.
In a second compression step, a more conventional zip-like compression is applied, which is called Huffmann-coding. This step works on redundancy left in the data and makes up for another 20% data compression.
Therefore, a MP3 file consists of a sequence of frames, each containing 32 bit header information and an audio block, which is made of Huffman codes. In many cases this frame is 1152 bytes long.
Watermarking techniques
There are several ways to embed the secret information into a music file. The technique to use depends on differerent criteria, like perceptual transparency, watermark bit rate, robustness, security, and computation cost. By deciding how to weight those criteria, certain algorithms pop up, while others don’t match those criteria.
One heavy criterium may for example be the need to keep the watermark even after a certain number of re-encoding or even digital-analog-digital transfer processes. To apply with that, a algorithm must hide the information deeply inside the sound data, which can easily lead to perceptable artifacts. The same is true for a huge number of bytes to be hidden.
But if , one the hander, one just wants to embed information on the receiver of a 320″, there’s no need to use one of those bulletproof, computational complex methods. Instead, changing the least significant bit (LSB) of certain elements in the data stream to store fragments of the secret text is a sound way, and there’s a number of (free) implementations.
All these methods, however, share one drawback, as they need to take action during the encoding process. There’s no (simple) way to embed data into an mp3 without re-encoding, because of the two-fold compression used and stored in the file. If you’d mess around with the Huffman codes, you’d generate seriously perceptable artifacts.
Consequently, to embed a watermark on the fly, like before sending a 320″ over AIM, you have to re-encode the original WAV. On modern computers this shouldn’t be much of a concern, taking just about 10 seconds.
And just to let you know: We do use watermarking! Hope you do so, too, soon!

After the issue lately has come up a couple of times in several forums, some words on the principle of watermarking may be helpful. Digital watermarking describes the process of embedding additional data into some digital document, like author or copyright information, without decreasing the document’s value. Taking it one step further, you wouldn’t want anybody besides you to see that embedded data, which is achieved by using a special watermarking technique called steganography, meaning hiding information within a digital document. I’ll use the term watermarking synonymically for both the general and the hidden approach thereafter.
MP3
Speaking of audio documents, aka tracks, the document type in question is mp3, as this is the common exchange format. MP3, or more specifically MPEG-1 Layer 3, as it is correcty called, is a lossy compression technique and file format, based on two internal compression steps. The first one takes into account certain properties of the human auditory system (HAS) to discard a certain amount of the original data.
When we hear a loud signal of a certain frequency, a second signal of similar frequency but lower amplitude taking place at the same time is masked, meaning we hardly perceive it. In a similar way, there also are timing masking effects. The MP3 encoding algorithm uses these properties of the HAS to reduce the stored data, while trying to keep the same perceptable audio quality. The user can ultimately determine this degree of perceptable quality by choosing a certain bitrate, which expresses the number of bytes the encoding process can use for storing a certain time interval of the source audio stream. By this, the audio data is packed into little pieces of data, called frames, which as a sequence make up the MP3 data stream. This also allows for streaming audio data on the internet as in IP radio.
In a second compression step, a more conventional zip-like compression is applied, which is called Huffmann-coding. This step works on redundancy left in the data and makes up for another 20% data compression.
Therefore, a MP3 file consists of a sequence of frames, each containing 32 bit header information and an audio block, which is made of Huffman codes. In many cases this frame is 1152 bytes long.
Watermarking techniques
There are several ways to embed the secret information into a music file. The technique to use depends on differerent criteria, like perceptual transparency, watermark bit rate, robustness, security, and computation cost. By deciding how to weight those criteria, certain algorithms pop up, while others don’t match those criteria.
One heavy criterium may for example be the need to keep the watermark even after a certain number of re-encoding or even digital-analog-digital transfer processes. To apply with that, a algorithm must hide the information deeply inside the sound data, which can easily lead to perceptable artifacts. The same is true for a huge number of bytes to be hidden.
But if , one the hander, one just wants to embed information on the receiver of a 320″, there’s no need to use one of those bulletproof, computational complex methods. Instead, changing the least significant bit (LSB) of certain elements in the data stream to store fragments of the secret text is a sound way, and there’s a number of (free) implementations.
All these methods, however, share one drawback, as they need to take action during the encoding process. There’s no (simple) way to embed data into an mp3 without re-encoding, because of the two-fold compression used and stored in the file. If you’d mess around with the Huffman codes, you’d generate seriously perceptable artifacts.
Consequently, to embed a watermark on the fly, like before sending a 320″ over AIM, you have to re-encode the original WAV. On modern computers this shouldn’t be much of a concern, taking just about 10 seconds.
And just to let you know: We do use watermarking! Hope you do so, too, soon!













Download vcard