Whats the most efficient way in Kyma to remove DC offset?

Question

Especially in a high replication count situation.... maybe the HPF is too heavy, and perhaps subsonic 10 hz or DC baseline shift could be removed more efficiently?

Answer 1

Another approach is to look at why there might be a DC offset in the first place.  One of the most likely culprits is samples or wavetables that do not end on zero.

If your original samples/wavetables do end and start on zero, try the Remove DC option in the Modifiers of the Wave editor to try to balance your original samples and wavetables.

When replicating, another trick you can try is to alternate the phases of the copies.  This can help prevent DC offsets if they're all the same sign.  To alternate the phases, you could put a Level onto the Input to the Replicator and set the Right and Left fields to alternate between 1 and -1, for example:

(?VoiceNumber mod: 2) * 2 - 1

Answer 2

The best way to remove a DC offset is with a High pass filter.  Set the cutoff frequency to something very low like 10 Hz so it won't affect the audible portion of your input, but it will attenuate or completely filter out DC, even if your DC is slowly changing.

Another approach is to take the difference of your input with itself delayed (for constant inputs, such as DC, the difference with itself delayed is 0 so it will cancel any constant signals).  Select a long enough delay so it doesn't affect the audio signal, though (the Difference functions as a high pass filter).

You could try applying the filter after the Replicator, but if the DC has already caused the Replicator to clip then it's necessary to apply the filter prior to the Replicator.  Another thing you could try is to attenuate the Replicator input to the point where it is no longer clipping; then high pass filter the Replicator to remove DC; then boost the levels back to where they were with a Gain.

Let us know how it goes!

Comments

is the HPF order relevant? I mean at 10hz should it be an order of 4 or could we get DC filtering also with an order of 1. This is much more efficient then, and better for replication purposes.

---

Yes, the order is relevant.  To determine how strong the filtering effect has to be in order to cancel the DC, you can experiment and see how it interacts with your particular input or set of inputs.