De Clicker

plugin_id: declicker · UI: De-Klicker (Impuls-/Click-Entfernung) · Source: plugins/declicker.py

Removes short impulse-like clicks/pops using a pragmatic detect → group → inpaint pipeline. It is designed for offline post-processing of stems (especially residual).

Parameters

Parameter	Type	Default	Range	Meaning
mode	choice	highpass	highpass	derivative
smooth_ms	float	1.0	0.05..10 ms	Moving-average window used only in highpass mode.
sensitivity	float	8.0	(typ.) 3..15	Multiplier for the robust scale estimate; higher detects more (risking transients).
min_level_db	float	-45.0	(typ.) -80..0 dBFS	Absolute minimum click level; prevents over-triggering on very quiet signals.
max_click_ms	float	5.0	≥ 0 ms	Skip segments longer than this (assumed not a click).
pad_ms	float	0.20	≥ 0 ms	Expand repaired region around the detection.
merge_ms	float	0.50	≥ 0 ms	Merge detection gaps up to this duration into one region.
mix	float	1.0	0..1	Blend repaired signal with original.

Algorithm

1) Build a high-frequency proxy

• derivative mode:

  proxy = |diff(x)|
  

  (expanded to also cover neighbor samples)

• highpass mode:

  lp    = moving_average(x, win = smooth_ms)
  proxy = |x - lp|
  

2) Robust thresholding (per channel)

sigma = 1.4826 * median(|proxy - median(proxy)|)    # MAD → σ estimate
thr   = max( median(proxy) + sensitivity*sigma, 10^(min_level_db/20) )
mask  = proxy > thr

3) Segment formation & repair

• Merge mask gaps up to merge_ms into a single segment.
• For each segment (s,e):
  • expand by pad_ms
  • if segment length exceeds max_click_ms: skip
  • otherwise repair by linear interpolation between boundary samples

x[s:e] = linspace(x[s-1], x[e+1])

Implementation notes

• The repair method is intentionally simple and fast; for large defects you usually want a more advanced restoration method.
• If strong transients are being “repaired”, reduce sensitivity, increase min_level_db, or switch to derivative mode.

References

• S. J. Godsill, P. J. W. Rayner, Digital Audio Restoration, Springer, 1998. (Click/impulse restoration overview.)
• F. R. Hampel, “The Influence Curve and Its Role in Robust Estimation,” Journal of the American Statistical Association, 69(346), 1974. (MAD as robust scale basis.)