SonauditA click is short, abrupt, and easy to describe after you hear it. Detecting one from samples is harder because many legitimate sounds have the same broad shape. A close-miked snare, a pick attack, consonants, vinyl texture, and a deliberate edit can all produce a sudden sample-to-sample change.
That ambiguity is why Sonaudit presents possible clicks and pops as listening candidates. A candidate is a request to inspect a moment, not a claim that the audio is damaged.
What a transient detector can observe
The current detector looks for a short run of unusually large changes relative to local sample-change statistics. It rejects changes that continue too long to resemble an isolated click, merges nearby candidates, and requires meaningful signal around the jump. Those rules reduce obvious noise, but they do not reveal the artistic cause of a transient.
A test signal can prove that a known synthetic click is found at the expected time. It cannot prove that every similar event in music is unwanted. The validation suite also includes a continuous high-frequency true-peak signal that once produced repeated click candidates; the detector was changed so that periodic signal is no longer treated as a series of isolated clicks.
Review the sound around the marker
Select the candidate and play the surrounding audio, not just the exact sample. A discontinuity between otherwise steady regions is more suspicious than a sharp event that aligns with a visible drum hit or consonant. Compare the left and right channels when the event appears in only one side.
Use the waveform as a locator rather than a verdict. A click can be audible without looking dramatic at the current overview scale, while a steep musical transient can look alarming and sound completely natural. Headphones and a quiet listening level are useful, but the final judgment still belongs in the context where the material will be delivered.
- Confirm the finding when the event is unintended and audible in context.
- Ignore it when the marker clearly belongs to intentional programme content.
- Leave it pending and add a note when the source or edit decision needs another reviewer.
The same caution applies to other content-dependent checks
Hum analysis can confuse sustained musical energy near 50 or 60 Hz with electrical noise. A negative stereo correlation can be intentional, yet still warn of mono-fold-down risk. Long silence may be an edit mistake, room tone, a dramatic pause, or required head and tail spacing.
Sonaudit keeps these findings in Review unless a more deterministic rule applies. A file that is effectively silent can fail; an internal silent range normally asks for listening. This distinction prevents a batch from being rejected solely because the content resembles a defect detector's positive example.
When a restoration tool is the better next step
Sonaudit locates and records candidates but does not remove clicks. If a click is confirmed, open the source session or a restoration editor where you can inspect individual samples, spectral context, crossfades, and alternate takes. Repairing a rendered delivery without the source may hide the symptom while leaving the underlying edit unresolved.
After the correction, rescan the file. The useful outcome is not a lower candidate count by itself; it is a reviewed delivery in which confirmed defects are resolved and intentional transients have an explicit decision.