Content-Adaptive Musical Audio Watermarking based on the Music Theory of Chords Arjun Yogeswaran, Pierre Payeur, and Jiying Zhao School of Information Technology and Engineering University of Ottawa Ottawa, ON, Canada [ayoge099, ppayeur, jzhao]@uottawa.ca Abstract—This paper proposes a novel content-adaptive music watermarking technique which uses the principles of music theory to enhance the capacity and robustness of watermark embedding. Using the musical concepts of key and chords, certain notes, which are musically coherent with the work, can be added or removed without impacting the listeners experience. The notes serve as the carrier of the watermarked bit. Since these notes are still within the human hearing range, with high enough strength, they will still remain intact after various audio compressions or distortions. The scheme does not require the original work to extract the watermark and serves as a framework to involve musical theory in optimizing watermark embedding algorithms. Keywords-audio watermarking; music watermarking; music theory; chords; content-adaptive. I. INTRODUCTION Digital watermarking has become a widespread technology, immersing itself into many different types of digital media including audio and video. Digital watermarking is the concept of altering a work to embed a message. Invisible or inaudible watermarking is the concept of imperceptibly altering a work to embed a hidden message. Watermarking has become very useful in the digital medium for many reasons, but most dominantly for protecting intellectual property rights, especially for the digital audio medium and the distribution of music. Many digital audio watermarking techniques make use of the analysis of the human auditory system (HAS) to hide the watermark [1,2,3]. Many techniques also use adaptive watermarking to modify the embedding method to more effectively embed the watermark based on the characteristics of the audio [4]. This paper proposes a new content-adaptive algorithm, designed specifically for music, which exploits musical theory to more effectively embed robust and imperceptible watermarks. The proposed method uses frequency analysis to determine the chords that are present in the music, and then proceeds to add or remove notes that belong in that chord to carry the watermark. By altering notes that belong in the chord, the correctness of the audio is not altered, and the listener should not experience any discomfort. Though there may be a slightly noticeable difference between the original and the watermarked work, the quality of the watermarked work will remain high and the listener should not notice that it has been altered. It is robust since compression and other distortions would not likely remove the embedded note due to its perceptibility from a human auditory system (HAS) standpoint. For detection, this technique does not require the original work, and uses side information to extract the watermark. II. AUDIO WATERMARKING There are many different techniques for digital audio watermarking which are generally classified into time-domain methods and frequency-domain methods. Time-domain methods generally embed and extract the watermark in the time domain. Two significant techniques that fall under time-domain methods are least significant bit embedding and echo hiding. Least significant bit embedding alters the least significant bits of each audio sample to embed the watermark [5,6]. On the extraction side, the least significant bits, containing the watermark, are read. Though it may be imperceptible, it is not very robust since any simple distortion can alter the least significant bits, and the watermark may be lost. Echo hiding is a method which is robust and remains quite imperceptible. The principle of the technique is to add a slight echo to frames of the audio [7]. This echo is of low amplitude and fits the audio so it is not likely to be noticed, however it can be easily identified by attackers [8]. The concept of frequency-domain methods is that the embedding and extraction of the watermark is done in the frequency domain. Phase coding embeds bits in the frequency domain by altering the relative phase of certain frequency components [9]. Slight variations in phase have proven to be imperceptible to the listener. However, compression algorithms may alter the phase and destroy the embedded data. There are also frequency-domain techniques that use very high frequencies and very low frequencies to carry binary data. This is useful for imperceptibility since the human ear is less sensitive to those frequencies, however compression algorithms will definitely degrade them since audio compression is usually based on HAS. III. MUSIC THEORY Musical theory is a mathematical language, defining the rules in creating music, and aiding in the analysis of music. 978-1-4244-4218-8/09/$25.00 ©2009 IEEE 129