AudioWatermarker<T>

Models & Types · AiDotNet.Safety.Watermarking

Keyed spread-spectrum audio watermark: a key-derived +/-1 sequence added to the 1-7 kHz block-DCT coefficients at a level that follows each frame’s loudness, detected by correlation.

Inheritance: ModelBase<T, Vector<T>, Vector<T>> → SafetyModuleBase<T> → AudioSafetyModuleBase<T> → AudioWatermarkerBase<T> → AudioWatermarker<T>

For Beginners

Audio watermarking hides a faint, password-protected hiss under the sound, shaped to stay below what you can hear. Anyone with the password can measure it, even after noise is added or the audio is re-saved; without the password it looks like ordinary noise.

How It Works

This is direct-sequence spread-spectrum watermarking (Cox et al. 1997, applied to audio by Kirovski and Malvar 2003) on the frames described in AudioWatermarkerBase. Per frame the amplitude is 0.2 x strength x max(spread, 0.03 x frame RMS), with spread the robust scale of the band’s coefficients: at the default strength the mark sits about 20 dB below the band’s own content. Detection normalises the received coefficients by their robust spread, clips them to +/-3 and correlates with the key’s pattern, which gives a z-score near N(0, 1) on audio without the mark. One second of 16 kHz speech-like audio typically reaches z of about 9, so the mark is found with full confidence; with another key, z stays near 0.

References:

  • Cox, Kilian, Leighton, Shamoon. Secure Spread Spectrum Watermarking for Multimedia (IEEE TIP, 1997)
  • Kirovski, Malvar. Spread-Spectrum Watermarking of Audio Signals (IEEE TSP, 2003)

Constructors

Constructor Summary
AudioWatermarker(double detectionThreshold, double watermarkStrength) Preserves the original constructor signature using the default configuration.
AudioWatermarker(double detectionThreshold, double watermarkStrength, string key) Initializes a new audio watermarker.

Properties

Property Summary
string ModuleName { get; }