Activated Content provides a robust, flexible and secure watermarking technology that embraces the new digital information age as well as established analog technologies. After reviewing the inadequacies of existing watermarking techniques, Activated Content's research team began its development with a fresh perspective. They integrated fundamental audio engineering principles with Psycho-Acoustic Compression Technology in a manner that had not previously been contemplated. The result is the Activated Audio Watermark-a patent-pending combined self-synchronizing and key based data carrying technology that fulfills all the criteria necessary to meet today's copyright protection requirements:
An Activated Content Watermark is inaudible.
It survives all forms of compression, encryption, transmission, duplication and conversion.
It is invisible to all forms of audio analysis.
And it can not be removed without damaging the original audio.
The Inaudible Watermark: The Activated Audio Watermarking system analyses the audio to reveal positions or locations in the signal that can hide the digital codes that are to be inserted without affecting the sound quality of the original. Using an advanced psycho-acoustic model, this analysis takes into account the signal masking characteristics of the human auditory system plus the effect of signal delivery systems and compression techniques on the original audio signal.
The Activated Audio Watermark falls below the threshold of audibility at settings tested and verified by "golden ears" from Major Record Labels and leading Mastering Studios around the world. Formal testing of the Activated Audio Watermark began in the fall of 2000 when a major Record Label conducted the first set of comprehensive double blind tests for audibility.
Historically, studies show that most people are naturally biased in the direction of thinking they can hear artifacts within audio, even when they are not present. Therefore, the particular design of data collection employed for the initial tests allowed not only for measurement of the audibility, but also of the bias and consistency of the observers.
Testing participants included an audio engineer, a digital sound mastering engineer, and a software engineer. All participants, in addition to the sound being listened to, had on display in the room a continuous spectrum analyser output, a stereo position output (Lissajou figure display), and a level meter.
The result of this formal double blind test is that there is no objective evidence that the watermark is audible in any of the tracks tested.
The Survivable Watermark: In order to survive the harshest delivery methods and most aggressive compression algorithms, the Activated Content watermark is designed to be inserted only in the regions of the audio signal that are ultimately delivered. Because this area of the audio spectrum is also the most sensitive for human hearing, special characteristics for the coded signals are needed to ensure that they are inaudible but still reliably detected by the decoders.
The special characteristic of Activated Content's coded signals withstand all compression techniques, like MP3 and Dolby coding and decoding, delivery media distortions like time and frequency variations, format conversions, manipulation of the audio through equalization, and copying. These characteristics also make detection of the watermark information reliable even within noisy audio. Once an audio signal is watermarked by Activated Content, it remains watermarked and the information remains detectable.
Numerous independent tests have been conducted by users of the Activated Audio Watermark. Early in the development of this new technology, survivability tests were conducted by Cambridge Consultants in the UK to verify robustness. Since then, users have consistently verified these results.
Cambridge Consultants encoded a watermark into a 45 second audio file - an excerpt from "Better Dreams", by "H." This file was chosen as past experimentation had proven this content to be a difficult example to watermark because the nature of the spectral content of the file gives little scope for modification without affecting audibility. Therefore a watermark embedded in this file should be somewhat fragile and susceptible to various common manipulations and distortions to the audio. It is Cambridge Consultants' belief that this is a representative 'worst-case' example for watermarking.
The watermark encoded was a 66 bit data packet. This was embedded twice in the segment of audio. The encoded file was then subjected to the distortions defined by the SDMI Phase I screen for robustness, and after each distortion, the decoder was used to recover the watermark. Samples of the results are given below. Please note that in the table below, the first 'bits correct' column indicates the number of bits recovered with confidence greater than 1 in 10000; the second 'bits correct' column indicates the number of bits recovered correctly, whether above or below this confidence.
The Invisible Watermark: Activated Content unique encoding technique effectively weaves the Activated Content watermark into a signal's audible spectrum by subtly altering minute fragments of existing sound in such a way that the watermark displays no acoustical characteristics or signature of its own. There is no tell-tale signature that can be identified through conventional audio analysis techniques. Because of this, the codes are not detectable and therefore undefeatable.
The Indelible Watermark: Because Activated Content's encoding technique seamlessly interweaves data into the audio signal, it is impossible to remove without seriously distorting the original audio signal. If anyone manages to even find the code-which in itself is improbable-tampering with the code will damage the audio substantially, greatly diminishing its enjoyment value.
Overview of Technical Operation: The Activated Audio Watermarking system consists of a key generator, an encoder and a decoder. The key generator produces a field of pseudo-random white noise, based on a 64-bit integer seed value. The encoder marks a music stream with data using the key, and the decoder reads back the data from a marked stream, again using the key. The same key must be used in decode operations as in encode operations or the encoded data will not be visible.
There are multiple benefits to a key-based data carrying watermark. The most obvious benefit is security. In applications whereby authentication of content may be somewhat sensitive a key based algorithm ensures that only those with access to the key file have the ability to decode and audio stream.
Additionally, this type of system creates the ability to define multiple watermark "name spaces" which can be defined by the user. In practice, watermark namespaces can be used to add multiple watermarks to a single audio stream (layering) or to partition watermark values by different customers, products etc.
The Activated Audio Watermarking system encodes watermarks of any bit length. Watermarks carry no semantic information, beyond that imposed by the application of the technology. This makes Activated Audio Watermarks extremely flexible and allows them to be repurposed for yet unforeseen applications.
The Activated Audio Watermarking System is file format agnostic as it operates on raw PCM data for both encoding and decoding. The encoder which processes unmarked music first converts it into an internal spectrogram format, encodes the watermark and then converts it back into the time domain. In the decoder marked music is converted into the same internal format before being analysed. In addition, the encoder and decoder make use of an error-correction algorithm to improve the reliability of data recovery while formatting bits are used to enable the self-synchronizing nature of the decoder.
Performance & Data Insertion Rate: Running on a fairly commonplace desktop PC (Intel P4 2.0+ GHz) and at its optimal settings, the encoder marks an audio stream at approximately 8x real time. At these settings, the encoder is able to insert approximately 1 bps, with slight variations based on the source audio (such as periods of silence).
Decoding time is non-deterministic as it is based on the quality of the source audio. However, real world experience shows that decoding rarely takes more than real time.
