What do BER and confidence mean in a detection result?
BER is the fraction of the 48 codeword bits that differ from the expected codeword in verify mode. Confidence is a normalized matched-filter score, not a probability. In blind mode BER is always 0 because the hard decisions are compared with themselves; use confidence, CRC validity and the z-score there.
How Aureal handles it
The detector correlates the audio with the key-derived reference codeword and forms a z-score from the aligned soft symbols. Confidence normalises that score to 0–1; it is a summary of signal evidence, not a probability that a person or owner is correct. A result counts as detected only when the CRC matches, the expected ID matches in verify mode, the confidence clears the threshold (0.5 normally, 0.35 with a passing CRC) and the recovered amplitude is positive. BER compares the hard bit decisions with the reference codeword, so it is meaningful only when an expected ID exists. The CRC-16 catches most errors but is not a cryptographic integrity check.
Common failure mode
“A confidence of 100% means the match is certain, and a BER of 0 proves the copy is perfect.”
Why it fails: Confidence is a capped normalisation of a correlation score, not a probability of authorship. In blind mode BER is 0 by construction, so it says nothing about quality. A CRC-valid codeword shows a consistent signal, not a perfect copy or a truthful claim about the file.
Verified example
Read the result in this order:
1. crcOk and recoveredPayloadId
2. confidence and z
3. ber only when you supplied --id (verify mode)
4. reps, Eb/N0 and SQNR for margin
A valid CRC plus the expected ID is the match. Report BER as signal quality, not as proof of anything.Example only. Validate against your own audio, codec and processing chain.
Scope and limits
- Blind-mode BER is always 0; it is not a quality metric there.
- Confidence is a normalized z-score, not a probability.
- CRC-16 detects many errors but is not cryptographic.
- A passing threshold is a signal result, not an authorship finding.
Frequently asked questions
Why is BER 0 in blind mode?
Without an expected ID, the hard decisions are compared against themselves, so no bits differ. Use confidence, CRC validity and z instead; BER becomes meaningful in verify mode when you supply the expected ID.
What confidence is good enough?
Detection needs at least 0.5, or 0.35 when the CRC passes. A marginal pass is still a pass, but record the margin (confidence, z, reps) so a later reviewer can see how close it was.
Related topics
Test the actual delivery chain. A fixture result is a starting point, not a guarantee for your material.
Open the Studio →