DataHiding^TM for Audio

The IBM DataHiding^TM technology for digital audio is a digital watermarking technology based on signal processing, probability theory, and the psycho-acoustic model. It allows record companies and studios to embed highly inaudible, robust, and secure signals directly into DVD- and CD-quality digital audio data streams such as PCM 96/48/44.1 kHz, 24/16 bits, and 6 channels/stereo/mono. The embedded watermark is detected in the uncompressed linear PCM domain. The degree of manipulation for embedding is held to be sufficiently small to be inaudible to the listener, but sufficiently large to yield a very low false positive error ratio when the Copy Control Information (CCI) is detected, and to show high ability to survive commonly employed or well-known processes of transmission, compression, filtering, and signal conversion. This is why the technology should be based on probability theory and the psycho-acoustic model in addition to signal processing.
This technology has been certified as excellent and practical for business purposes at STEP 2000 and STEP2001, an international evaluation project for digital watermark technology for music in 2000 and 2001.

The technology can realize any predetermined low false positive ratio at the time of detection while maintaining high transparency if an appropriate threshold and time for the detection window are set. The false positive error ratio of the submitted detector in detecting the CCI data is less than 10 for each 15 seconds of the detection window. Deliberate unauthorized removal or updating of the embedded data is extremely difficult without serious degradation of the audio quality.

• Flexibility in optimizing the overall trade-off
  The data structure of a watermark allows flexible optimization of the trade-off among the competing requirements of transparency, data capacity (payload), detection reliability, survivability, and security,
• Multiple layers of watermarking
  Up to three layers of watermarks can coexist without interfering with each other.Unrelated watermarks can be embedded at different times and locations.
• Low false positive error
• High transparency
  High transparency is achieved on the basis of a tuned psycho-acoustic model.
• High survivability
  • Two successive D/A and A/D conversions
    The embedded signal survives the sequence of the processes D/A, A/D, MiniDisc-recording, MiniDisc-playback, D/A, A/D.
  • FM radio broadcasting
    The watermark can be retrieved after successive D/A, frequency-modulated radio broadcasting, reception by a consumer receiver, and A/D processes.
  • Resampling and down conversion of channels
    The watermark can be detected at any sampling rate down to 16 kHz.
  • Time compression/expansion with pitch shift and pitch preserved
    Linear speed changes within 10% and pitch-invariant time scaling within 4%
  • Data reduction coding
    MPEG-1 L1/L2/L3, MPEG-2 AAC, and ATRAC-1
  • Nonlinear amplitude compression
  • Additive or multiplicative noise
    For example, the embedded signal survives white noise addition with a constant level of 30 dB lower than the long-term averaged music power.
  • Frequency response distortion such as equalization
  • Addition of echo
    For example, maximum delay: 100 msec, feedback coefficient: around 0.5.
  • Band-pass filtering; for example, 8 kHz low-pass filtering.
  • Flutter and wow: 0.5% rms., from DC to 250 Hz
  • Overdubbing.

• Internet distribution of music (one application of this is the EMMS)
• Counting of commercial film broadcasts
  How many times a commercial film has been broadcast can be automatically counted if its sound track contains an Audio DataHiding watermark in it.
  
• Playback counting
  Artists can find out how many times their music has been played on a particular TV or radio channel if they embed identification information into their music in advance.