Searches an audio file for clicks: hard cuts, single stray samples and places where the waveform suddenly changes direction. Every place comes with its time, how far it stands out from its surroundings, its level above 5 kHz and the fade length that would bury a click of that size. The file is read in the browser and not uploaded.
Choose an audio file
WAV and AIFF are read sample by sample; FLAC, MP3, M4A, Ogg and Opus go through the browser's decoder.
| Time | Kind | Channel | Stands out by | Click above 5 kHz | Fade that buries it |
|---|
A linear predictor of order 16, fitted anew for every 2048 samples, predicts each sample from the previous ones. Tonal material disappears in the prediction error, a jump remains. A place is reported when the error in its 1 ms block stands at least 12 dB above the louder of the two 20 ms surroundings and is at least 18 dB sharper than the error around it within 1 ms. The second rule separates a single outlier from a percussive attack, which spreads over many samples.
A jump is a sudden change of value, as with a cut in the middle of a waveform. A bend keeps the value but reverses the direction: in the measurement for the article on crossfades, a cut exactly at a falling zero crossing still clicked at −37.44 dBFS. The level above 5 kHz is the same measure as in that article.
The fade length comes from the same measurement: a fade of 1 ms lowered the click by 33.08 dB, and above that every doubling brought another 6 dB, up to 59.35 dB at 20 ms, independent of the pitch of the material. The tool picks the shortest length that lowers the click by its excess over the surroundings plus 6 dB.
ffmpeg's adeclick works on the same principle and replaces outliers by interpolation. At its defaults, in a dense test mix without a single click, it still changed 10.6 % of all samples, and afterwards this search found 18 places there instead of one; in a pad it removed 15 of 20 cuts and changed 1.4 % of the samples.
Articles on this
How loud the click at a splice is and which fade length buries it is measured in the article “Crossfades and Cuts”.
Sources: Image-Line: Sampler Channel Settings (Declicking) · ffmpeg Filters Documentation (adeclick, acrossfade)
In a dense test mix with sixteenth-note hi-hats, 7 of 20 hard cuts stood out and were reported; the rest were buried in the treble of the hats. Very hard intended attacks, such as a synthetic kick that starts with a jump, can be reported as well.