Does the watermark survive background noise?
The local fixture matrix records additive noise at −30 dBFS for all three bands, and the README's adversarial table names a −20 dB pink-noise fixture. Both are local results. Recovery depends on the signal-to-noise ratio inside the carrier band, not only on the file's overall loudness; broadband noise and 8 kHz paths are weak hosts.
How Aureal handles it
Spread-spectrum detection integrates repeated frames, so noise that looks severe on a waveform can still leave a decodable mark when enough frames are available. The relevant quantity is the in-band SNR at the carrier, which is why a quiet recording with low self-noise can outperform a loud one with broadband hiss. The embedder also mutes the watermark in near-silence (at or below about −54 dBFS) to keep pauses clean, so a clip made only of room tone may carry no mark at all, regardless of its duration.
Common failure mode
“A −20 dB noise floor is fine because the fixture passed at −30 dBFS.”
Why it fails: Overall dBFS and in-band SNR are different measurements. Pink noise concentrates energy at low frequencies, while the carrier sits in the mid or high band; the result also depends on how many frames the file provides and on any codec that follows. A level that passed on one fixture says little about a different noise spectrum.
Verified example
Measure the result at two noise levels on the actual master, and keep the carrier band in the record:
ffmpeg -i marked.wav -f lavfi -i "anoisesrc=color=pink:amplitude=0.01" -filter_complex "[0:a][1:a]amix=inputs=2:duration=first" noisy.wav
node aureal-watermark.cjs detect noisy.wav --id 883921
Repeat at a higher amplitude and compare confidence and BER rather than a single pass/fail.Example only. Validate against your own audio, codec and processing chain.
Scope and limits
- The noise results are local fixtures, not platform guarantees.
- Broadband noise is a known weak host.
- The embedder mutes the watermark in near-silence (≤ about −54 dBFS).
- Codec damage after the noise compounds the loss.
Frequently asked questions
What SNR do I need?
There is no single number that holds across content. Detection integrates repeated frames, and the useful margin is the in-band SNR at the carrier. Test the actual noise spectrum and codec, and record confidence and BER.
Does noise erase the ID?
It raises the bit error rate and can push a marginal file below the detection threshold. With enough frames and a strong enough mark it may still decode; the fixture results show the direction, not a guarantee.
Related topics
Test the actual delivery chain. A fixture result is a starting point, not a guarantee for your material.
Open the Studio →