Crossfade vs Separated Crossfade

[IchiOtoSasayaki]

Hi,
Do you know what the difference is in these situations?

In the left image #1: At the top, the orange channel has a crossfade. Below it, in light blue, is a copy of this channel with inverted phase, resulting in them nulling completely.
In the right image #2: I dragged the right part of the orange channel's crossfade to a new channel, and now they do not null at the crossfade.

Thank you

[Hugh Robjohns]

Oh dear!

Do they no longer null just through the cross fade, or after the cross fade?

[IchiOtoSasayaki]

Just momentarily through the crossfade.

But just for the sake of accuracy, even in image #1 on the left, depending on the type of crossfade, there was sound at the crossfade at around -145 dB, so I'll consider it a null.
But in the second example, depending on the type of crossfade, I sometimes get a momentarily audible sound at -55 db for the same crossfade just separated.

[Hugh Robjohns]

Okay, so it's down to the maths of the reciprical crossfade curves and the summation of multiple channels. Sounds like something's not adding up!

Which DAW is this?

[IchiOtoSasayaki]

This is Cubase 10.5, I haven't updated it yet. Maybe it has already changed in newer versions...

[Wonks]

The 2nd one’s just two fades, not a dedicated crossfade. The crossfade dialogue seems to have more control.

Is it a real problem or just an experiment?

[Hugh Robjohns]

-145dBFS is the rounding error level of the 24bit output, so that's as good as it gets and working perfectly.

-55dBFS is not going to be a problem in most practical applications, but does suggest a weird problem somewhere.

Clearly, the summation of two channels works properly since the nulling works.

So is it an issue with summing three channels, or is there something odd happening when you separate the fade in / fade out?

Intriguing!

[Hugh Robjohns]

The 2nd one’s just two fades, not a dedicated crossfade.

Ah... in that case, all bets are off. Two different fade curves/durations are unlikely to null perfectly.

[IchiOtoSasayaki]

The 2nd one’s just two fades, not a dedicated crossfade. The crossfade dialogue seems to have more control.

Is it a real problem or just an experiment?

Theoretically, I would expect them to be the same since they have the same unchanged parameters, but it does seem that the dedicated crossfade is calculated differently! I'll ask Steinberg about it and check the newer version.

Yes, it may be a problem for me, as it is part of a technique I'm working on in sound design involving layering. If I want to apply different effects to different parts of the crossfades as channel inserts instead of directly on the audio event, I would prefer to trust that when separating the audio, it remains unchanged.
I will need to work more with this technique to know how much of a problem it really is as it is.

[IchiOtoSasayaki]

-145dBFS is the rounding error level of the 24bit output, so that's as good as it gets and working perfectly.

Good to know about the -145dBFS!

[Hugh Robjohns]

Theoretically, I would expect them to be the same since they have the same unchanged parameters, but it does seem that the dedicated crossfade is calculated differently!

I tried the same crossfade-and-drag-to-a-second-channel thing in SADiE and it nulled perfectly through the crossover.... so it's a weird cubase anomaly!

I want to apply different effects to different parts of the crossfades as channel inserts instead of directly on the audio event, I would prefer to trust that when separating the audio, it remains unchanged.

Unlikely to be an issue if you're processing the tracks anyway.

[IchiOtoSasayaki]

I tried the same crossfade-and-drag-to-a-second-channel thing in SADiE and it nulled perfectly through the crossover.... so it's a weird cubase anomaly!

Thanks for checking in another DAW, good to know that it could work perfectly!

Unlikely to be an issue if you're processing the tracks anyway.

Thanks, I hope indeed!

[Wonks]

I've just repeated this experiment on Cubase Elements 13 and can get similar results, but if I use a really short crossfade (1/16 bar overlap at 120 BPM) I get a bit less noise in the middle (about-65dBFS peak on the meter). With 1 bar overlap at 120 BPM I got around -55dBFS, but at a 3 bar overlap I got an indicated perfect cancellation on the meter.

I'm trying to test it out on Studio One but there isn't a simple polarity reversal option for editing the audio and the polarity switches on the groups don't seem to have any effect!

[IchiOtoSasayaki]

Thanks Wonks for confirming this on the newer version!.
Now I don't need to do it myself. I'll inform Steinberg about it, and that they've been beaten by SADiE.

[Wonks]

It must be down to the maths used. Maybe 32 bit float vs 64 bit float.

With the crossfade, you can calculate one gain value then take that away from the unity gain value to get the gain for the second slope. So no reason not to match up.

With two separate slopes, there's independent calculations going on. You're bound to be converting figures to log values and back again, so the more accurate the calculations, the better the cancellation will be. But for normal fades with non-identical waveforms, being very slightly out really isn't an issue.

[Wonks]

And I've now managed to get it working in Studio One 6.5 Pro and it stays fully cancelled when you split the first track into two tracks with fades.

So Cubase could do better here.

[Hugh Robjohns]

That's good to know. I was sure SADiE wasn't the only DAW that can do proper sums. Shame on Cubase!

I'd be surprised if Reaper messes it up. Anyone tried yet?

[Drew Stephenson]

It's on my list!

[James Perrett]

I'd be surprised if Reaper messes it up. Anyone tried yet?

I've just tried and it appears to depend on the crossfade shape. The linear fade and the S shaped curves are fine but the default curve (the second one down on the list) appears to not cancel. I think it may only 3dB down at the crossing point rather than 6dB. However, I've been using Reaper for years so it may be set up to use a very old default.

[Drew Stephenson]

Interestingly the linear fade on the example I've just tried nulls completely as a cross fade but not in the three-track example shown above. It's down at -88dB but not zero as it is when it's a crossfade on a single channel.

[sonics]

Perhaps in Cubase the fades are not perfectly aligned with the overlap?

It all works perfectly in Studio One, but I can replicate the issue by moving the fades just a little, which makes sense. (To see the phase buttons in SO you need to enable the "Input controls" visibility in the Channel Components menu)

Have you tried it using a different timebase to see if that affects it?

[Wonks]

In Cubase (as in Studio One), with the crossfade on the top track, you can simply drag the second section down with snap on, and the crossfade becomes a fade out and a fade in.

I zoomed in and the fades seemed to be in the right places, but there’s what you see and what the program actually works on, which could be slightly different things.

[Wonks]

Have you tried it using a different timebase to see if that affects it?

As I mentioned in my test, I got different results with a 3-bar fade (full null), a 1/4 bar fade (-55dBFS) and a 1/16 bar fade (-65dBFS). So it wasn’t consistently getting worse with shorter fades.

[IchiOtoSasayaki]

Thanks everyone for testing, very interesting results.

Have you tried it using a different timebase to see if that affects it?

I have tried in Cubase with different "Ruler Displays": Bars+Beats/Seconds/Timecode, with and without quantize, with and without zero crossing, 32bit float and 64 bit float, and results seem to be the same.

[RichardT]

Interesting - I agree the best thing is to talk to Steinberg. On the face of it, I can’t see why these wouldn’t get much closer to nulling.