k y m a • t w e a k y — the kyma collective || view the current website || February 2015 Archive

/ WebHome / Sounds / Share.DescribeWienerNR

Search


Share Section


Sounds
Samples
Timelines
Tools
Microsounds

Home 
Topics 
More... 

All Sections


Products
Order
Company
Community
Share
Learn

Login / Register 
Change password
Forgot password?

Symbolic Sound


Home
Kyma Forum
Eighth Nerve

TWiki Links


twiki.org
This Soundfile contains several elements. The basic one is a different kind of Noise Gate, most useful for very low level thresholds. Then that element is combined into one channel of the Dolby style spectral processing, and finally a complete Stereo Spectral Processor is provided built up from these elements and the previous precision limiter and soft clipper.

Wiener filtering is a process by which you can derive the optimal reconstruction of a noise contaminated signal. From optimal filtering theory, you can show that at any frequency the optimal filter has amplitude given by

FilterAmpl? = 1 - Noise^2/(Signal+Noise)^2

Generally this is used in conjunction with high resolution spectral decompositions as you get with Fourier Transforms. But in this case, where we have 5 bands of processing, we have enough spectral decomposition to do a pretty reasonable job of applying the Wiener Filtering concepts.

The Kyma compressors/expanders are primarily devised to work at moderate to loud signal levels. At low signal levels their table interpolation in amplitude space is too coarse. To overcome these effects for handling gating at very low signal levels (e.g., < -60 dBFS), the Wiener gate provided here works quite well.

Also included is a kind of spectral compressor = high postgain compression at low thresholds + original signal, to produce a curvilinear kind of compression used to enhance low level musical details without crushing the peaks of stronger signals as you would have with typical compression. This one is based on direct computation of the dB level, applying linear compression in dB space, then converting back to amplitude space. These conversions are performed using 3 term Minimax polynomial approximations over limited domains, and with scaling to cover the rest of the amplitude space.

The result of the Log2(x)/32 is to compute dB(x)/192. This is then used here to produce something that is unity gain below threshold, and strong limiting above threshold. The result is therefore always either gain neutral or attenutation. That is good because the 2^(32*x) routine only works for values of x below zero to produce answers less than one.

One of the problems you encounter when applying strong amounts of spectral compression processing is that you end up magnifying the noise in the recording. So the Wiener gate is provided to help overcome some of that effect. Be careful when setting the Noise Floor (NF) estimates in each channel. At and below this level in dB, there will be no transmission of the signal at all. If the NF is set too high, then you will end up stuttering as the signal hovers around that level. It is better to accept a little bit of noise content by setting NF intentionally lower by several dB. You can estimate the needed NF levels by watching the channel meters with NF set to -100 dBFS while playing a looped section of background noise at the tail of a recording. Setting NF to -100 dB gets the Wiener gate out of the way.

This same technique is the basis for many noise cleaning operations, although in those cases a much higher spectral resolution is generally employed.

-- DavidMcClain - 21 Mar 2004

 
 
© 2003-2014 by the contributing authors. / You are TWikiGuest