DeFilter Suggestion for Simple "HDR" DeFiltering

Official support for:
17 posts since 17 Dec, 2017

Post Mon Sep 02, 2019 9:13 am

First off, when I say HDR I'm referring to HDR tonemapping in image editing where you attempt to fit a large range of values within a smaller range by compressing it down by different values over an area. Adjusting contrast over different frequencies / blurs / areas of pixels so that fine details and edges are retained but the overall image is darkened or brightened to a smaller range.

With DeFilter at 20ms filter length it captures virtually every little frequency, attempting to flatten it out into noise. But what if we blurred that and made more gradual changes to the overall timbre, which shorter filters seem to have issues with. I experimented with linear blurring by frequency shifting with feedback and delay and also experimented with logarithmic blurring by frquency modulation / vibrato, I think log blurring works best. What this does is smear the sound spectrally so that all the little peaks and valleys that would get detected and evened out into noise are unheard by DeFilter and remain in the signal when learning is stopped and the blurring disabled, giving the sound a flatter overall tonality but with more of the original timbre retained.

Currently, without using external plugins, the best way to do this is to use a shorter filter length but they seem to always have issues in the high end and don't look smooth at all.

If this was built into DeFilter then the user could decide how granular to get with the spectral flattening, either cancelling out every small detail or if it should look at the overall shape of the sound. Perhaps either as a single slider for the level of blur, or perhaps several sliders controlling a frequency range of the blurred spectrum so that you can target the very broad shape, leave the "medium" peaks and valleys, but also flatten the sharp peaks and valleys. Perhaps a really wide one called Tone or Slant or Slope, a less wide one called Structure, a sharper one called Timbre, and a really sharp one called Tonal.

Also, a built-in target EQ profile of Pink Noise or just "Pink" would be nice, I find Music to be very dull. ... _forth.mp3

Here's a short demo of a breakbeat going through DeFilter with a 20ms filter targeting pink noise, first without filtering, then with filtering, then with blurred filtering, then back and forth to make it more obvious. Without blurring it sounds more like bursts of noise, but with blurring some of the original, albeit comb-filtery, timbre is retained. Sounds with better timbres would have clearer benefits. ... 8/Amen.mp3

Here's the Amen break, 20ms filter, 3ms filter, 20ms blurred filter, same EQ after the blurred and non-blurred DeFilter since they both had the same tonal issue of overblown bass and toneless snares. Listen to how the characteristic rides are turned to noise without blur, but with blur they retain their timbre, all with the accuracy that a longer filter length allows. Here's the filter at 20ms 3ms 20ms blurred Notice there's still peaks and valleys in the 3ms larger than the 20ms blurred filter, which isn't necessarily bad but implies that it's targeting finer details than the blurred filter, even when the filter length is very low.

I'm also not convinced that vibrato / FM on the detected signal is the best way to do it, there might be something else. This is just what I had available and was easy on my CPU. Perhaps vibrato that fades out as the pitch increases or decreases so it's almost gaussian weighted. Alternatively, and this would require something else entirely, something like a Median filter could remove tonal outliers that would just bloom out into peaks and valleys if it was blurred.

User avatar
966 posts since 1 Jan, 2005 from Norway

Re: DeFilter Suggestion for Simple "HDR" DeFiltering

Post Sun Sep 15, 2019 11:07 pm


Thanks for your thorough testing and the suggestions! This is definitely within the scope of a future DeFilter update. I'm not sure if the "constant Q" smoothing you describe is the best approach, though. The psycho-acoustical frequency scales like Mel or Bark usually give better results.


Return to “Acon Digital”