CEDAR Duo – declickle and auto dehiss
20
Declickle tutorial
Detect
Your first job is to choose the correct detection mode for the material to be restored. Listen to
the audio, and decide whether it contains well-defined clicks, whether it suffers from grungy
clicks and crackle, or whether the problems lie somewhere between. When you have done so,
select the fine, coarse, or medium modes (respectively) as you see fit.
The best results for a given piece of material will be obtained when - if possible - you use a finer
mode. So, for example, if you feel that the problems lie between 'coarse' and 'medium', you
should in preference try to find settings that work with 'medium'. Likewise with 'medium' and
'fine'... try to use 'fine' whenever possible. This will ensure that the high frequency and transient
content of the wanted audio is best preserved.
Threshold
Once you have chosen the detection mode, you must find a suitable value for the threshold. You
can often leave this at its default value of 10, reducing it only when removing fine crackle and
buzzes, or increasing it if you notice any dullness or distortion.
If you reduce the threshold too far, the wanted audio may exhibit mild damage in the form of a
slight instability in background sounds and any background noise present. You should always
attempt to find the highest value at which the unwanted clicks and crackle are removed.
The correct threshold will vary with the detector mode chosen. Fine mode will require a lower
threshold than coarse mode if it is to detect the same number of clicks and crackle.
Model
Once you have selected the detection mode and threshold, you must find the most appropriate
value for the modelling. For most purposes, you can leave this at its default value of 99, which
allows Declickle to remove clicks and crackle accurately in the presence of difficult signals such
as trumpets and violins.
However, if the audio contains a high-frequency buzz (regularly spaced clicks) the harmonic
modelling within Declickle will recognise this inappropriately as wanted signal. and attempt to
ensure that this is not removed. In this case, you should reduce the modelling parameter to a
suitable value. Always use the highest value that removes the problem. This will minimise the
risk of damage to the high frequency content of the genuine signal.
