Level differences between midi and audio versions of same tracks

[Markdu]

As the title says. I like to mix in audio rather than midi so I usually record all midi tracks onto their own separate audio tracks. I do this in Logic. The setup involves VEPro on a PC slave which hosts all my orchestral instruments, currently Hollywood Orchestra/Play. To bounce down the tracks I simply set the output of each midi track and the input of each audio track to the same bus and then record to them in real time, as I find off-line bouncing doesn't work properly with this setup. This works fine except for the fact that sometimes there is a difference in level of up to 1dB between the midi and the 'host' audio tracks. Often only in certain regions and not others on the same track and not on all tracks. The difference is not very audible, to my ears anyway, but it does show up on the meters. So I am wondering why this happens when I am simply recording the output of the midi tracks straight to audio. Should't they be exactly the same including in terms of level?

[fuzzfilth]

Instruments often use Round Robin assignments, which means a row of identical notes of the same velocity will still trigger different samples which will result in slightly different levels. That's the point actually.

 

For perspective - please do ask an orchestra to do a second take and may they please stay within 1dB of level difference. I'd pay money to see their faces.

[David Nahmani]

For perspective - please do ask an orchestra to do a second take and may they please stay within 1dB of level difference. I'd pay money to see their faces.

 

Interestingly, the same would be true for many patches on synthesizers, whether they're analog synths, hardware or software digital synths. Most patches have enough non-linear elements (unsync'ed oscillators, random modulation sources, unsync'ed LFOs etc....) that would not playback exactly the same sounds (and thus the same levels) twice in a row.

[Markdu]

Instruments often use Round Robin assignments, which means a row of identical notes of the same velocity will still trigger different samples which will result in slightly different levels. That's the point actually.

 

For perspective - please do ask an orchestra to do a second take and may they please stay within 1dB of level difference. I'd pay money to see their faces.

 

Ok thanks for the perspective! Actually, yesterday one or two of the recordings ended up significantly more than 1dB different, to the point where I will probably need to re-do them. Some of the others too.

 

Yes, I think the round robin aspect could well explain some of this and thank you for reminding me. I do check all the recordings afterwards so obviously I am re-playing the midi versions which will sometimes use different samples because of that. However that only explains some of the behaviour as a lot of the instruments/articulations don't have round robin assignments, e.g. basic sus patches etc. I will check the tracks that are the most problematic though to see if they do use them.

 

So if this doesn't explain all this behaviour, are there any other factors that could be at play here?

[Markdu]

For perspective - please do ask an orchestra to do a second take and may they please stay within 1dB of level difference. I'd pay money to see their faces.

 

Interestingly, the same would be true for many patches on synthesizers, whether they're analog synths, hardware or software digital synths. Most patches have enough non-linear elements (unsync'ed oscillators, random modulation sources, unsync'ed LFOs etc....) that would not playback exactly the same sounds (and thus the same levels) twice in a row.

 

Didn't see this until after I posted that reply. Yes, fair point. However most of the tracks do stay true to their levels and also, apart from the round robin samples, there isn't a lot within the instruments that I am aware of that would vary much if at all between play-throughs. Otherwise every time I played the piece, or worked on it (the midi), it would sound significantly different which it doesn't, even if I play it repeatedly from the middle somewhere. Also, previous projects I've worked on with the same setup didn't have this issue, at least not in a way that stood out for me.

 

So at this point I'm not convinced that it can all be explained by the built-in variability of how the instruments work.

 

Edit: I just looked at this again and one instrument that uses a simple sustained patch with no round robins plays repeatedly at one point with the midi track at a level of -35.2 dB and the audio track it was recorded to at -32.7. That's nearly 3 dB difference and quite audible. (It's obviously a very quiet part of the arrangement at those levels!)

Edited March 11, 2021 by Markdu

[David Nahmani]

It's hard to tell without knowing the instruments and more specifically the actual patches you are using. But if you want to experience what we're talking about, create a software instrument track, choose a Drum Kit Designer patch, program a simple drum beat and create a 1 or 2 bar cycle. You're bound to hear something different on each pass of the cycle.

[Markdu]

Sorry David, missed your post again! Please see my edit above. The instrument is a Cimbasso from Hollywood Brass Gold and the patch is called 1CM Sus Non Vib if you happen to have that library. Thanks.

 

I should add that the mod wheel was used to create a swelling note so that CC data is involved in the 'performance' on each pass. However I didn't set CC #11 to anything at that point, though the volume (CC #7) is set at 0 dB.

[Markdu]

A bit of an update, though I seem to be talking to myself at this stage (was it something I said? )

 

I've noticed that the instruments that are stored on my Mac, i.e. not hosted by VePro on the PC, record with the levels exactly the same. They also use Play though are from a different library and are loaded directly into Logic. Most of the instruments which are hosted in VePro are coming out with these level differences, some worse than others. That seems like an important factor, though I've yet to discover why since the outputs of both the midi and the audio tracks are in Logic and that's exclusively where I mix.

 

Hmm...

[robinloops]

If you’re working with them after you commit to audio, why not just make appropriate adjustments on the mixer faders to adjust to the new protocol (now audio files instead of midi) and be done with it? Or can put a gain plug-in on each track (I use a vu meter vst with a gain knob for this function) and make the adjustment there. Only has to be done once for each track and is minimal (time) compared to bouncing offline and would only add seconds to the process of ‘preparing’ each track for audio.

[Markdu]

I should have been clearer that it is only certain regions on any given audio track that end up with different levels as compared to the midi, usually not the whole track. So fader or gain adjustments that affect the whole track wouldn't work. Plus some of these orchestral instruments end up with slight timbral differences as well which can really alter the sound, albeit quite subtly. So I've now ended up with a bunch of audio tracks, parts of which will need adjusting or re-recording to match their midi counterparts. Which shouldn't be the case if, like with the instruments that are hosted within Logic itself as opposed to in VePro on the PC, they were simply 'copied across' to audio exactly as they sound in midi. Something is getting in the way of that some of the time and I have established that it is not a round robin issue as many of the affected instruments aren't using that. I really need to get to the bottom of this because the amount of re-adjustment needed is proving to be pretty unfeasible.

[Mania]

What is the point mixing everything to perfection in midi then turning everything to audio expecting it will be the exact same mix? Maybe just stay with MIDI?

 

Or: just don’t perfect everything in midi since you know you will turn everything into audio later and have to MIX again. Just do a rough balancing with midi so you have a song and feeling. When it feels good and you know it has POTENTIAL, just bounce everything so you can start the real MIXING.

[Markdu]

Thanks but my point is it should be the exact same mix. There's nothing that I can see that should make the midi performance turn out in any way different once it is rendered to audio. Staying with midi is what a lot of people do when working with orchestral samples but on my system it very quickly starts to glitch and overload once I start adding reverbs, delays etc. as I do the final mix. Also, many of the characteristics of the instruments have to be managed in midi using mod wheel, expression, velocity and other CC's so I have to get each performance right in midi before rendering it. Which is why it is problematic if the audio is significantly different. The final mix adds effects and adjusts final levels etc. but the initial performances all have to be perfected in midi beforehand and to stay consistent once rendered to audio.

[Mania]

I see. Well it looks like you don’t have too many options left if you can’t figure out how to get the exact print out for audio that you had in midi.

Have you tried freezing the tracks once they are OK? This way you’re left with more CPU power.

Other than that you probably just have to accept the fact that after exporting to audio you have to do further mixing. Good luck and let us know if you figure out what was causing your issue.

[Markdu]

Thanks Jim. It's infuriating because I used to get the exact same in the audio tracks, for some reason that has changed and no, I can't figure out why! Freezing certainly saves on resources but the way I learned to mix orchestral material was based on the free manipulation of audio tracks and that is by far my preferred method for a number of reasons. Well, as you say I might just have to put up with it as it is but I hate unsolved mysteries so I will certainly try to get to the bottom of it!

 

I appreciate your and everyone else's input.

 

Mark

[Mania]

And the obvious: when bouncing to audio: do NOT check normalize?

[Markdu]

Yes absolutely. In fact I bounce all these in real time, not in place, because doing the latter with the slave system ends up with notes cutting off and other artefacts, so normalizing is not even an option!

[Markdu]

Actually one thing I could try come to think of it is freezing the tracks first and then rendering them to audio. That might preserve the midi exactly as it is supposed to sound. I know the frozen tracks are in fact audio tracks but they are not available to edit in the same way as ordinary audio tracks. Might try that tomorrow..

[David Nahmani]

(was it something I said? )

Nooooo! I'm just real busy at the moment.

[Markdu]

(was it something I said? )

Nooooo! I'm just real busy at the moment.

 

No problem, thanks for chiming in anyway!

[robinloops]

Only some regions... that would be an issue. Also more quantifiable as an issue and not just variation between passes.

[Markdu]

I don't think it's down to variation between passes as I can play over the same midi region multiple times and still get the same level, as well as other characteristics. If I record one region to audio in real time it seems to come out the same. However the problem seems to be when I bounce multiple tracks at the same time to audio, again in real time. That's when I get these random level changes.

 

I also forgot that I cannot freeze tracks with this setup (Logic on the Mac, VePro on the slave system hosting the instruments). I get the same thing as I do when trying to bounce in place, i.e. most of the notes, especially long ones, get cut off after the initial attack. So my idea of freezing the tracks first and then bouncing to audio is not a runner.

[David Nahmani]

I've noticed that the instruments that are stored on my Mac, i.e. not hosted by VePro on the PC, record with the levels exactly the same. They also use Play though are from a different library and are loaded directly into Logic. Most of the instruments which are hosted in VePro are coming out with these level differences, some worse than others.

Ok, so my earlier point about trying a 2 bar loop with Logic's Drum Kit Designer was to demonstrate that some instruments/patches (but not all) use some kind of random element that makes them play slightly different upon successive playback. My guess is, that could be what you're experimenting that results in a level difference.

 

I don't think it's down to variation between passes as I can play over the same midi region multiple times and still get the same level, as well as other characteristics. If I record one region to audio in real time it seems to come out the same. However the problem seems to be when I bounce multiple tracks at the same time to audio, again in real time. That's when I get these random level changes.

Ah, ok. Then how exactly do you bounce multiple tracks at the same time? Which menu/function do you use?

[Markdu]

Ah, ok. Then how exactly do you bounce multiple tracks at the same time? Which menu/function do you use?

I create the number of audio tracks that correspond to the midi tracks for all the instruments of the orchestra. I do this procedure section by section so as not to overload the system. The midi tracks have the VePro plugin on them but no actual instruments - those are on the slave. I also have an incident of Scripter on each midi track which runs a 3rd party script which applies all CC curves, i.e. modulation, expression etc. to all channels within a single incidence of Play so that all of each instruments' articulations, which are set to different channels within that instance, can be affected by the same CC data. That's the basic setup.

 

Then I set each midi track's output to an available bus. I use the same bus as the input for its corresponding audio track. I record-arm each of those audio tracks and then press record so the piece plays through in real time recording the output of each midi track to its own audio track.

 

I hope that gives you a clear idea off how I do it. I'm certainly open to a better way if there is one..