Joining frequencies from two models

 

Sometimes it may happen that a regular ensemble even with min spec doesn't give you complete freedom over what you want to achieve, having one cleaner narrowband model result with a fullband model result with more vocal residues, but you still want to have a full spectrum.

 

Instead of using ensemble Colab, you can also mix in some DAW, MDX-UVR 464/inst3 or Kim inst model result which have 17.7Hz cutoff, with HQ_1-5 or Demucs 4 result, which has full 22kHz training frequency model.

 

First, import both tracks. Now rather the most correct attitude to avoid any noise or frequency overlapping is to use brickwall highpass in EQ at 17680Hz everywhere on Demucs 4 stems, and leave MDX untouched, and just it. You can use GSEP instead of Demucs 4 (possibly less vocal residues).

 

If you want to experiment further, as for a cut-off, once I ended up with 17725.00 flat high pass with -12dB slope for "drums" in Izotope Elements EQ Analog and I left MDX untouched. “Bass” stem set to 17680.00 in mono and "other" in stereo at 17680.00 with Maximiser with IRC 1 -0.2, -2, th. 1.55, st. 0, te 0. But it might produce hissy hi-hat in places with less busy mix or when hi-hat is very fast, so tweak it to your liking.

 

For free EQ you can use e.g. TDR Nova - click LP and set 17.7 and slope -72dB.

As a free DAW you can use free Audacity (new versions support VST) or Cakewalk, Pro Tools Intro, or Ableton Lite.

 

The result of above will probably cause a small hole in a spectrum, and a bit lack of clarity. Alternatively, you can apply resonant high pass instead of brickwall, so the whole will be filled without overlapping frequencies.

 

Instead, you can also consider using linear phase EQ/mode like in free Qrange and its high pass to potentially cause less problems in phase.

 

A similar method to this can also be used for joining YT Opus frequencies above 15750Hz with AAC (m4a) files, which gives more clarity compared to normal Opus on YT. Read this.