Most Transparent Multiband Splitter

[btfnk]

I use Fabfilter MB for this. Mostly to build multiband Ableton racks.

[Funkybot's Evil Twin]

Anyone try Blue Cat’s MB-7? Not sure what it uses for crossovers but it can host VSTs in each band. Always meant to check it out.

[highkoo]

I use KiloHearts Multipass a lot-
https://kilohearts.com/products/multipass
I dig it a lot, but I cant say Ive ever really compared it to FF MB, which is probably the only other thing Id compare it to..

[fairlyclose]

I use this https://www.bluecataudio.com/Products/Product_MB7Mixer/

[Aesaire]

Haven't tried FIR/FFT filters before so I can't comment.

I see that you are a Reaper user. Sake's NOTT JSFX effect has a linear phase mode and LR. It is not a crossover though, but a multiband upwards compressor.

Think I have that already, haven't played around with it before, will check it out later.

Got quite a few more recommendations, will probably run some tests when I have the time.

EDIT: By the way, if you do a phase inversion test on a sound source with crossover filters, would it be accurate to say that the quieter the end signal (with absolute silence aka a null being completely transparent), the more transparent the crossover filter?

[rafa1981]

The problem is how to define "transparent". Audible or measurable?

This is an hypothetical case/nitpick; Let's say that we have:

1- A linear-phase filter with some imperfection on the magnitude response that is audible on an AB test.
2- A minimum-phase (non linear phase) filter with perfectly flat magnitude response and the traditional phase shifts that are not audible on an AB test.

It is still perfectly possible theoretically, for filter 1 to score better on the null benchmark. It will most often do unless it is very wonky, but for that case the filter causing less audible impact is the second one.

In practice I guess that yes, the better it nulls the most transparent, but at the end let your ears be the judge.

[Autobot]

::::

[Aesaire]

The problem is how to define "transparent". Audible or measurable?

This is an hypothetical case/nitpick; Let's say that we have:

1- A linear-phase filter with some imperfection on the magnitude response that is audible on an AB test.
2- A minimum-phase (non linear phase) filter with perfectly flat magnitude response and the traditional phase shifts that are not audible on an AB test.

It is still perfectly possible theoretically, for filter 1 to score better on the null benchmark. It will most often do unless it is very wonky, but for that case the filter causing less audible impact is the second one.

In practice I guess that yes, the better it nulls the most transparent, but at the end let your ears be the judge.

That's a good point. The null test gives a very theoretical measure of transparency, when in reality it ignores the material dependent nature of the sound. Maybe comparing crossover filters is a futile exercise if it always comes down to use your ears and see which one suites the material better.

I’ve recently discovered that a lot of multiband operations (dynamic and saturation for example) are better in single band with compensated emphasis or creative sidechain

I'm also not such a fan of splitting a signal into multiple bands and prefer other operations like sidechaining and smart filtering at certain positions in the signal path. I also rarely have the need for that kind of stuff because I do not often work with sounds going over the whole freq spectrum.

That said I use EQ8 in Live to split the signal. If I measured it right there is no harm to the signal at all. Nulls perfectly and no phase issues. What you see is EQ8 Live (no oversampling) with a 48db four band split. SPAN and Bertom EQ Curve Analyzer used.

EQ8 is just the stock equalizer in Ableton though right? Could you detail the steps of your test or provide a project file? I'm extremely skeptical that a 48db four band split would perfectly null. No crossover filter I've tested comes remotely close to this from linear phase to minimum phase.

Only thing remotely similar when using ISOL8, where if the multichannel mode or mute/solo band buttons wasn't engaged it didn't actually start the crossover filter. Thought it was perfectly transparent before that.

[FapFilter]

48db is incredibly steep though. Noone would expect to get natural results using these while EQing and i think 48db filters and more are more like the last resort you can reach for if you have really bad material that that can't be fixed otherwise and thus needs to get rescued somehow and steep filtering is still the lesser evil.
Or if you are purposely going for it as a creative effect.

Of course it'd be nice if everything could be perfectly isolated with 100% precision while being totally transparent, but i guess it's probably just prevented by the laws of physics

[Ploki]

Yep. I try to keep it at max 18dB/8ve

[Jbravo]

I think FLStudio recently added a native splitter. In the past I've just use a multiband compressor (with compression disabled) so there are probably more available tools than people realise.

[NAD]

There's this trick for creating a 3-band crossover you can do with some clever routing if your DAW happens to support it:

• Send the signal through a low-pass filter to one channel for processing the low frequencies
• Send the signal through a high-pass filter to another channel for processing the high frequencies
• To a third channel send the full range signal and also send the high- and low-passed signals but invert them so you end up with only the mid frequencies

From there you just buss those three channels together or group them in a folder track and obviously don't forget to disable the master send on the original track.
I got this idea from a fantastic Dan Worrall video in which he uses linear phase filters although even if you use any other type of EQ this still works and will null with the original signal.

[Autobot]

::::

[Aesaire]

There's this trick for creating a 3-band crossover you can do with some clever routing if your DAW happens to support it:

• Send the signal through a low-pass filter to one channel for processing the low frequencies
• Send the signal through a high-pass filter to another channel for processing the high frequencies
• To a third channel send the full range signal and also send the high- and low-passed signals but invert them so you end up with only the mid frequencies

From there you just buss those three channels together or group them in a folder track and obviously don't forget to disable the master send on the original track.
I got this idea from a fantastic Dan Worrall video in which he uses linear phase filters although even if you use any other type of EQ this still works and will null with the original signal.

Wow, this works. I wonder why no one has tried turning this into a plugin.

It is only possible with FIR/FFT crossovers, but you should expect them to have latency.

So it is either phase shift (not a problem if you aren't mixing with the dry signal) or latency.

https://www.youtube.com/watch?v=951MnO8M1Qs

Is this an implementation of the FIR/FFT crossover you were talking about? Subtracting the highs and lows from the mids through phase inversion offsets many artifacts in a crossover filter.

[rafa1981]

ReaFIR uses FFT convolution, so yes. This is exactly the case.

Notice that on IIR filters, other than on the single pole case, substracting leaves the bands with different phase responses, so the substracted bands don't have a flag magnitude response. If unprocessed they reconstruct perfectly when summed back though.

Subtracting only works as intended on linear phase filters.

On MixMaxtrix 1.1 I'm adding a subtracted IIR crossover with selectable number of poles, but just for creative purposes and for the 1 pole version, which is almost perfect. I already nailed that bug.