Audio Watermarking Denis Lebel presented by

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 2 / 13 Presentation Outline Introduction Introduction Systems Systems Techniques Techniques Applications Applications Discussion Discussion References References

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 3 / 13 Introduction Idea: Embedding an inaudible mark into an audio signal Idea: Embedding an inaudible mark into an audio signal Originally proposed as a technique to counter music piracy Originally proposed as a technique to counter music piracy Analogical to the technique used for paper (e.g., money) Analogical to the technique used for paper (e.g., money) First audio watermarking techniques were directly inspired from previous research on image watermarking First audio watermarking techniques were directly inspired from previous research on image watermarking

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 4 / 13 Introduction Fingerprinting vs Watermarking Fingerprinting vs Watermarking Fingerprinting analyzes the signal and constructs a unique signature Fingerprinting analyzes the signal and constructs a unique signature  Signal is not modified  Requires a repository  No preprocessing required Watermarking hides information in the audio signal Watermarking hides information in the audio signal  Signal is altered  Self-contained  Signal must be preprocessed

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 5 / 13 Watermarking Systems Devices or systems should check for watermark before proceeding with operations Devices or systems should check for watermark before proceeding with operations A detection mechanism is required A detection mechanism is required A key (as in cryptography) is generally used during construction and detection A key (as in cryptography) is generally used during construction and detection Symmetric (same key) Symmetric (same key)  Both construction and detection use a private key Asymmetric (different keys) Asymmetric (different keys)  Construction uses a private key and detection uses a public key Watermark is permanent. Audio data can still be used. Watermark is permanent. Audio data can still be used. Contrasts with encryption: temporary and content only usable when decrypted Contrasts with encryption: temporary and content only usable when decrypted

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 6 / 13 Watermarking Systems Properties Properties Inaudibility:No sound quality degradation Inaudibility:No sound quality degradation Robustness:Resistance to ANY signal transformation Robustness:Resistance to ANY signal transformation Capacity:Bit rate Capacity:Bit rate Reliability:Error rate during detection Reliability:Error rate during detection Low Complexity:Efficiency Low Complexity:Efficiency

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 7 / 13 Watermarking Systems Psychoacoustic models are often used to ensure inaudibility Psychoacoustic models are often used to ensure inaudibility Figure 2: Watermarking shaping example. (Kim 2003)Figure 1: Masking curve example. (Kim 2003)

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 8 / 13 Techniques Spread-Spectrum Spread-Spectrum Spreads pseudo-random sequence across time-domain signal or transform signal Spreads pseudo-random sequence across time-domain signal or transform signal Pseudo-random sequence is generated using a secret key Pseudo-random sequence is generated using a secret key Watermark is embedded as a modulation of pseudo-random sequence Watermark is embedded as a modulation of pseudo-random sequence Watermark is scaled according to a psychoacoustic model Watermark is scaled according to a psychoacoustic model Figure 4: Spread-spectrum scheme. (Kim 2003)

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 9 / 13 Techniques Replica Replica Uses the original signal itself to create the watermark Uses the original signal itself to create the watermark Example:Echo Hiding Example:Echo Hiding  Introduce echo in time domain  Watermark bit values are embedded using 2 different delay values  Detection finds delay length used to determine watermark  Masking is used Could also be done in frequency domain Could also be done in frequency domain  Example: Frequency shifting

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 10 / 13 Techniques Self-Marking Self-Marking Embed special signal in the audio Embed special signal in the audio Example:Embedding a peak in frequency domain Example:Embedding a peak in frequency domain Two-Set Two-Set Statistical method based on hypothesis testing and relying on large data sets Statistical method based on hypothesis testing and relying on large data sets Pseudo-random process to insert a certain statistic into the audio signal Pseudo-random process to insert a certain statistic into the audio signal Usually applied in frequency domain Usually applied in frequency domain Example:Patchwork Algorithm Example:Patchwork Algorithm

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 11 / 13 Applications Copyrights Copyrights Proof of ownership Proof of ownership Enforcement of usage policy Enforcement of usage policy Forensic watermarking Forensic watermarking Fragile watermarking Fragile watermarking Fingerprint watermarking Fingerprint watermarking Information hiding Information hiding Added value Added value Annotation Annotation

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 12 / 13 Discussion Is watermarking really useful? What could it be used for? Is watermarking really useful? What could it be used for? What could be done for copyright protection? What could be done for copyright protection?

MUMT-611: Music Information Acquisition, Preservation, and Retrieval 13 / 13 References Gomes, L., P. Cano, E. Gomez, M. Bonnet, and E. Battle Audio watermarking and fingerprinting: For which applications? Journal of New Music Research 32 (1): 65–81. Gomes, L., P. Cano, E. Gomez, M. Bonnet, and E. Battle Audio watermarking and fingerprinting: For which applications? Journal of New Music Research 32 (1): 65–81. Craver, S., M. Wu, and B. Liu What can we reasonably expect from watermarks? IEEE Workshop on the Applications of Signal Processing to Audio and Acoustics. 223–6. Craver, S., M. Wu, and B. Liu What can we reasonably expect from watermarks? IEEE Workshop on the Applications of Signal Processing to Audio and Acoustics. 223–6. Kim, H, Y. Choi, J. Seok, and J. Hong Audio watermarking techniques. In Intelligent watermarking techniques, edited by J. Pan, H. Huang, and L. Jain, 185–219. River Edge, N.J.: World Scientific. Kim, H, Y. Choi, J. Seok, and J. Hong Audio watermarking techniques. In Intelligent watermarking techniques, edited by J. Pan, H. Huang, and L. Jain, 185–219. River Edge, N.J.: World Scientific.