Advanced Crossfeed DSP

http://www.changstar.com/index.php?topic=1054.0

That thread deserves to be copied over to this site, v3 is still the best software crossfeed I've heard, by a good margin. Forget Isone and bs2b.

 

Is AstralStorm still around? I haven't tried this yet though I use Jriver's standard crossed at home all the time.

 

I like this crossfeed too! The most natural for sure, but I just use only with fb2k because I can ajust the gain to 0 or -1 dB on Jriver vst menu doesn't open and I didn't like listen with -12dB default.

 

You shouldn't set gain over -3dB or so or it will cause clipping if the file is loud enough. Inevitable when mixing signals from 2 or more channels.

 

I am still extremely slowly developing it, but generally day job takes it over. Main thing I've been working on isimproved gain handling so that clipping should no longer happen - but it will get quieter. That version has a fixed lowpass which should remove one combing issue as well.
Additionally it has balance controls and bypass button.

The v3.0 is: https://www.dropbox.com/s/wsqlb9rkixpmzdy/adv_xfeed_v3.zip?dl=0
New source you can build with Faust ( https://faust.grame.fr ) is attached - the new faust build script from distribution will handle it too.

That new room speaker preset for Senn HD650 on my head:
Center boost: 4.5 dB
Corner frequency: 450 Hz
Gain: -7.5 dB
HF feed level: -6 dB
LF feed level: -1 dB
Head diameter: 17.28 cm
Speaker angle: 30 °

Different results if you set center boost to 6 dB, gain to -9 dB, freq to 3800 Hz, feed levels to HF -12 dB and LF -6 dB and angle to 60° - near field monitor-like.

 

Oh right, the main source of inaccuracy is in the highs, due to phase and/or frequency response of the fractional delay. If anyone has a good paper on even more accurate delay than this DC-accurate (L² or if you prefer least-mean-square optimal) Lagrange delay, let me know. The difference is single dB and single degrees of phase, but can be in some rare cases audible - typically as slight darkness.

 

can't wait to try this out!

 

Apparently I need to rethink the approach I was taking to further improve the plugin. It seems that whoever devised the equations I used made a few critical errors so the actual angle + width tunables are quite wrong.

I'll be working on devising the right ITD function and more advanced feed signal equalization. As it is, most of the effect is caused by the comb effect which can be tuned with this but not directly.

Center boost interacts with all the gain tunables quite a lot.

Example new tuning guide for BS2b style with reduced artefacts:
- set 0 degrees (disabling the trig equation for ITD)
- set some center frequency close above 1000 Hz (not critical - literature suggests 1500 Hz)
- set all gains to 0
- tweak manual ITD tunable up from 0 until width is ok and highs coloring is low - use a "wideband" track with a hard stereo pan sweep or hard panning
- reduce HF gain until width is correct
- increase main gain to match loudness
- tweak LF gain until bass is not in head - preferably use an inverted stereo bass track for this (chiptunes are most egregious here)
- tweak gain to match loudness again (likely slightly positive)

Use bypass switch often to check is loudness still matches.

Results for me:
Center frequency 1300 Hz
ITD 4.0 us
LF gain -1.5 dB
HF gain -6 dB
Main gain 1.5 dB

Next, you can tune the "distribution" of the center using Center Boost.
You will have to readjust gain to match loudness (see the theme here?), then change HF and LF gain as per previous guide.
After these steps check if imaging is correct on a track with precise location of sounds and reverberation - e.g. orchestra recording.

When done right this changes soundstage depth and shape only - high values flat, low values "scooped".
Seems there's a function of gains vs center gain too - because it does indeed reduce crossfed signal while also cutting reverberations due to reduction in normal side signal.

After the final tweak we get:
Center gain 0.5 dB
LF gain -0.5 dB
HF gain -3.5 dB
Main gain 1.0 dB
ITD 3.0 us

Then make sure mono signals stay in the center by fixing channel imbalance in your headphone. (Possibly with the help of a stereo eq.)

I suspect that ITD becomes too phasey at high frequencies adding combing - and the delay is actually the pinnae reflection which is supposed to vary with frequency. I'd have to look at time domain HRTF recordings and yet more research.

 

Hi AstralStorm, I love using your plugin. Any chance you’ve had a chance to update and further improve on this? Also, with jriver migrating to 64bit (I’ve held off upgrading), is your VST files 64bit compatible?

Thanks