In this article Michael Carnes, reverb guru and the man behind Lexicon and Exponential reverbs, shares his experience of mixing a solo classical piano recording to stereo and to Dolby Atmos. In a sparse recording where big panning moves aren’t appropriate what does Atmos bring to the project?

Mixing in Stereo

I was recently asked to record two pieces of piano music played by my friend Jason Hardink (pianist with the Utah Symphony).  In this article, I’m going to describe the process of making both a stereo mix and a Dolby Atmos mix. 

Mixing in classical music—and especially so in chamber or solo music—isn’t at all like the process of mixing popular music of any form. Mixing in pop (a generic term I use) is absolutely part of the creative process. Placement, EQ, effects are all part of that vocabulary.  Of course, you mix in classical as well, but not as dynamically.  We trust that the player got it right—dynamics, timbre and all. So our job is first and foremost not to leave any evidence we were ever there. Mixing for a solo instrument or small chamber group is generally pretty static: once you’ve got the balance, levels and positioning you shouldn’t really need to do anything.

It’s different for an orchestra. There’s no such thing as a ‘true’ vantage. The players are spread over a large space and no static balance is ever going to work perfectly.  There are spots (accent mics) that may need to come in and out. Reverberation levels may need to change as the music moves along. While you could make the argument that—even for solo piano—certain changes in balance might contribute something artistically,  I felt that Jason had that all under his fingers.

What do I want out of a stereo mix?

Some of the answers are obvious—clarity, musicality, reasonable mono-compatibility. Another thing that’s important to me is robust width.  I really do dislike the notion of the ‘sweet spot’. Of course, there’s a place where room calibration and speaker positioning are targeted. The mixer has to work there. But many mixes will collapse if the listener moves off the sweet spot. There must be decorrelated time differences across the plane so that people around the room have an enveloping experience.

I mixed in the same session that I used for editing.  The only difference is that I consolidated (rendered) the tracks so that they no longer had edits I might mistakenly pull apart. I am still working at 192K with all files in floating-point. 

With that, let’s look at the mix window and then a few of the plugins I used. The dark blue tracks on left are the tracks on disk.  They’re all routed to aux channels (purple) for mixing. This simplifies my converting this mix to Atmos. Here you can see that my A/B pair (flanking omnis) and room mics are hard left and right. These give me the robust width I need.  The “Decca pair” are spread just a little less and the close mid/side pair give me focus and detail, along with good mono if needed.

This is the mix window, showing all the tracks.  For most Pro Tools users, this will be a very simple mix. There’s one thing here you might find interesting. The M/S decoded output goes to the 5th channel (as seen in the graphic above). You should notice that I have left panned to right and right panned to left, with both channels toed-in. If you take a look at the video, you’ll see my M/S mics will actually pick up more of the treble strings on the left side and bass on the left.  This is not an exact low-to-high stereo image because strings in the piano cross.  But high notes predominated on the left side and low on the right.  I decided the listener would enjoy a vantage similar to the player, so I simply flipped the channels.  Panning hard to left and right gave me too much separation.  Toeing both sides towards the center gave me a more natural sound and also gave some of the mixing that comes from the piano lid.

Now let’s look at the plugins I used.

Voxengo MSED

M/S decoding is easy to do with just a bit of bussing in Pro Tools (or any other DAW).  But it’s even easier to do with a plugin.  This free decoder from Vogengo is a handy tool to have.  It takes a two-channel input (assumed to be mid and side) and gives you a decoded output.

iZotope Neutron 3

I’d mentioned that the room had a little whistle around 16K.  This was quite low and probably not hearable.  But it bugged me.  Rather than going after it in RX, I simply used a very tight and targeted notch. Neutron 3 gave me great visual feedback so I could see what I was doing.  Notice I did this in the individual tracks rather than the final mix.  That left some room for the reverb to fill any gaps (I doubt there were many, to begin with). The tamed version of the whistle is well under -100dBFS.

iZotope Imager

In the article on tracking the piano, I mentioned that I used an Omni pair spaced about 10” apart and off the tail of the piano.  This is a technique that was often used by Decca and gave a nice sense of spaciousness. The off-axis signal had a slightly greater spacing than ORTF, while the piano itself was pretty close to omni. It was hard for me to find a place for this pair in the mix. But I tried something that worked out pretty well. In iZotope’s Ozone suite there’s a little tool called the Imager. My best guess is that it looks for intensity differences between channels and converts that into a time difference.

Imager gave me a greater and more robust width while still leaving the central piano with only very minor time differences.  Mono testing gave me very satisfactory results.

Exponential Audio Stratus 3D

The hall itself sounds very nice, but I wanted to extend tails a little.  And RX can dry things up as well, so I wanted to correct for that.  Even though this is a stereo mix, I used Stratus 3D so I could make a preset that also worked in Atmos. The reverb level is. about 12dB down from the dry audio.

Mastering

Once I’d completed the mix, it was time for a little mastering. There are some dangers in doing your own mastering, so it’s important to take a breather at the end of the mix and put on a new hat. Since this was a project that was targeted toward video, I needed to make sure that the overall level (as measured in LUFS or LKFS) met generally accepted standards. I aimed for an integrated level of around -16LKFS. These pieces were both quite dynamic, so I needed some gentle limiting and compression to get there.

At the very end of the process, I use Insight 2, which measures power, imaging, max levels and so on. Over the course of the entire mix, I aimed for an integrated loudness between -16 and -17 LKFS. This lets me get the mix into a range that can work on desktop, tablet, phone while not sounding manipulated. Insight is a passive measurement tool: the actual adjustments are made in Ozone.

Ozone is a great tool that’s relatively easy to begin using.  in the EQ panel, I made a couple of very minor tweaks. In mastering, the smallest changes can make a very big difference. In the dynamics section, there was very gentle compression (1.7 : 1) below -10dB.

To finish up, I did an offline bounce to 48K sampling rate with a 24-bit word size.  This went over to Ashkan for final mixing into the video.  This converted the audio to AAC.  Unfortunately, lossy compression is the standard for video streaming, but by keeping the mix at the highest possible standard, we preserve as much as possible.

So that’s the approach for a stereo mix. What about Dolby Atmos?

Mixing in Atmos

Making an Atmos mix of classical music is great fun, with some necessary restraint. Some music is made to be manipulated: this could include most forms of popular music, film music or even some advanced electronic music. But most classical music is made to be heard in a recital space or concert hall. Flying a piano around the room is tempting, but generally not appropriate.  Our interest in mixing this music in Atmos is to give it a firmer sense of place in a way that puts the listener in a convincing hall.  In this mix, I opted to place the listener at the piano keyboard (or perhaps a step or two back).  I felt it was the best way to hear the music.

My mixing space is a pretty typical minimum room for doing Atmos. There’s a speaker directly in front (the center speaker), another pair in the traditional stereo location (L/R), another pair on either side of me (sides), and a pair in the back corners of the space (rears). I also have a subwoofer to give me accurate low frequency discernment. So far, this is a standard 7.1 setup.  I also have 4 speakers above me—one pair ahead and another behind. So the whole arrangement is called 7.1.4.  There are rooms with a few more speakers and some with many more. I wouldn’t mind a few more myself, but I haven’t yet found a tree that grows dollars.

I’m going to avoid a full description of Atmos here, but one of the real merits of the format is that it scales pretty well to what you have—whether it’s earbuds, a typical A/V surround system, a soundbar or a full-blown multi-speaker room. With all that out of the way, let’s take a look at how the mix was laid out:

We’re in Pro Tools, which is connected to the Dolby Atmos Renderer. The recorded audio is at the left, in dark blue. These four stereo channels are the original 8 microphone channels.  I’ve finally downsampled them to 48K, which is the only sample rate currently supported by Atmos. These channels are not routed directly to outputs.  Instead, they are sent to aux busses.

Panning of the original signal

The aux busses go to 5 channel strips (5-9 in purple). These strips are Atmos objects. From left to right, you’ll first see the strip receives the M/S signal.  In this strip, you’ll see M/S decode, along with some EQ, a bit of light compression and a limiter. This is an odd place for a limiter, but we’ll talk more about mastering a little later on. This decoded signal is cross-panned (as it was in stereo) and toed-in just a little. The objects are widened as well.

The next channel strip (#6) receives only the mid signal from the M/S pair. This object is routed to the center channel.  This gives a much firmer anchor than a phantom center, and the image will hold up much better as you move around the room. The rest of the processing on this channel is largely the same as the first strip, with a few small differences in EQ.

Moving to channel strip #7, we have the signal from the A/B pair of flanking Omnis. These are panned to the side speakers and the signal is widened a little. This gives a great sense of spaciousness without being obtrusive.  I added a very small bit of delay (3-4 milliseconds) to make sure that the Haas effect didn’t cause the listener to hear sound originating from the sides.

Channel strip #8 has the output from the Decca experimental pair, widened as in the stereo mix. These mics are panned midway between L/R and sides. In a larger Atmos room, this signal will actually be assigned to the “wide” pair of speakers. Either way, the placement provides a nice unobtrusive bridge between the M/S signal and the side signal.

Strip #9 contains the signal from my high room mics and is panned to the top speakers, midway between front and back.  This would normally go to the bed, but I’ve done a few things that I’ll talk about later on.

Panning into the reverb

You can see the way the reverb reacts. Inputs are in blue and outputs in green.

You should notice that each of channel strips 5-9 is also attached to a reverb send.  This send is the maximum width supported in Pro Tools, which is 7.1.2.  The panning into the send follows the same path as the object routing. For example, a signal panned to front center enters the reverb at front center. A signal panned to left side enters the reverb at left side. A properly-written reverberator will react in a natural manner to this placement.

Signal coming out of the reverb

The reverb signal is sent not to the Amos bed, but to a 7.1.2 bus.  The top corner channels are picked up in channel strips 11 and 12.  Only the low reverb channels (7.1) are sent to the bed. The high channels are sent to objects which are panned as shown in the image of the mix window above. You’ll also notice that the top center reverb (or top side as I should have labeled it) is also ducked by about 10dB relative to the other height reverbs.  In part this is because the room omnis are panned to the same location. But there’s more to it.  That requires a whole new paragraph:

The Atmos bed is problematic in the height channels, since there are only two high channels in the bed. While a very minimal Atmos playback system may only have two height speakers, most will have more. If you have only a single stereo height signal in your reverb, it will be smeared across all of the height speakers.  This actually results in a reverb that is less spacious. So let’s say you have 4 height channels in your reverb.  If you have 4 height speakers, the entire room will open up and the sense of reverb will be much more spacious.  But if you add the high bed channel, that spaciousness will degrade, since the bed is mixed into all of the height channels.  By ducking that channel by several dB, this deleterious effect is much reduced.  But we don’t want to through the baby out with the bathwater.  What if the playback system has 6 height speakers? If we’ve taken that bed pair and placed them into an object panned midway between high front and back, you’ll get 6 discrete reverb signals, which will sound really great. This folds down acceptably to 4 and 2 height speakers. This means that the height reverb should always go into objects.  Forget there’s a bed up there.

Minimize use of the bed

The Atmos bed is more or less  a crutch for those coming out of the stereo and surround worlds. Unlike objects, the signal is time-aligned, just as it has been in theaters for the last 30 years or so. But the bed can’t take advantage of a system with more speakers and it greatly lowers the positioning accuracy of your mix. 

That doesn’t mean objects are problem-free.  Objects are not time-aligned based on their position in a mix, so you must have a general notion of the room the mix will playback in. I suppose it makes me a teensy bit sad, but I think there’s not likely to be a theatrical release of the mixes I’ve described here. This means that, whenever necessary, I introduce alignment delays that will work in smaller rooms. A film mixer has to worry both about large theaters and intimate living rooms. My hat is off to those mixers—they are real pros.

Now back to this mix…

Taking a last look at the mix window (I’ll give you a minute to scroll back up to the top of the page), you’ll see three more channel strips. Strip #14 is simply the low part of the reverb going straight to the bed. Strip #15 is a VCA, which allows me to control the level of all signals on the way to the Atmos renderer.  There’s no such thing as a master strip for Atmos, so a VCA (or nested set of VCAs) is the cleanest way to manage overall level. And finally strip #16 is linear timecode going to the renderer to keep it in sync.

The Atmos Renderer

The loudness metering is very important. In Atmos music, the integrated loudness should normally be no higher than -18LKFS. I’m at -19.7, which is right about where this piece needs to be. The true peak should be no higher than -1dBTP. I’m at -1.3 and I won’t argue with that!

Mastering in Atmos

Mastering is—to me at least—the hardest part of Atmos. The creative part is fun, but when you’re delivering a mix, you really have to be very close to the specified loudness measurements. There are not really any mix-wide tools to help.  There’s no 128-channel compressor, limiter, EQ—anything really. Mono/stereo versions of those tools are still very helpful. I used them here in this quite simple mix. But there’s a lot of non-intuitive stuff going on in Atmos.  An adjustment of the VCA won’t necessarily give you a linear change in the overall loudness.  That really depends on where the energy is in a particular piece. Experience in a given genre makes a big difference. But until you’ve gained that hard-won experience, you can count on putting in lots and lots and lots (and lots) of hours.

Summary

I have to confess that it ended up taking a lot more time and required a lot more words than I envisioned when I decided to share my experience.  For me, it’s back to composition for a while.  I’ve assembled links to the videos as well as audio-mastered recordings right here. If you are a classical recordist—even a beginning one—I hope there’s something for you in this project. If you’ve been around the block a while longer, I hope you’ll enjoy moving a project into Atmos and finally if you want to read about all the steps in this project from recording to mixing you can find it here.