Mixing Stems

(Right off the bat, these are really rough notes and late-night thoughts on the topic while I'm waiting for 7-Zip to do its thing. This entire page needs to be updated with pictures and a proper tutorial at some point, but if I wait to do things "right", they'll never get done. Hope it's useful in the interim.)

Getting your stems together

So stems and multitracks are often confused with each other. Multitracks are the raw channels on an eight-, 16-, or more channel recording. This is the sorta thing that someone mixing a record would be working with, and they usually have much less processing applied to them. Stems are the multitracks after they've been folded down into instrument parts, and usually after some sweetening has been applied. You might have eight drum multitracks for the snare, hi-hat, kick, overheads, and so on, but only one drum stem (or perhaps three, if you're mixing for Rock Band).

So basically, stems are less unwieldy multitracks, still with separated instruments, but usually all the effects and mixing on them already baked in. You'll be creating them if you're not working with mixed audio.

GH2 uses a three-stem arrangement, generally speaking. For guitar/bass songs, you have a band-guitar-bass setup in the VGS. "Band" is anything not playable. Your count-in, drums, vocals, synths, effects, and background guitar parts will be on the band stem, while "guitar" and "bass" are naturally the playable instrument parts. Lead/rhythm songs use two VGS files, a singleplayer band-singleplayer guitar VGS and then a band-lead-rhythm VGS for multiplayer. (You need two VGS files because the singleplayer guitar plays a mixture of the lead and rhythm parts.)

The actual channel mapping differs from song to song, but here's what you'll find in the base game, usually:

  • Six channels (stereo band, stereo guitar, stereo bass, or perhaps stereo lead/stereo rhythm): The most channels you'll find in any GH2 VGS. Stereo bass is by and large a waste of space, unless you've got some heavy panning.
  • Five channels (stereo band, stereo guitar, mono bass): This is most of the songs in the game.
  • Four channels (stereo band, stereo guitar, or mono lead/mono rhythm): All lead/rhythm songs use this channel mapping for their singleplayer VGS. I know at least two songs in the game use a four-channel multiplayer VGS, "Soy Bomb" being one of them.

You can use any of these if you're setting your own channel mappings in songs.dta. If you're not, you'll need to make sure that your VGS matches the number of channels in the VGS you're replacing. Otherwise, the game will crash on song load.

Actually mixing audio is out of the scope of this tutorial, but basically, make sure you've got a band track, a guitar track, and a bass track, and you're good.

Loudness and dynamics

GH2 requires some of the most cooked audio I have ever heard for its stems. Perhaps with EQ, things could be done a little bit less aggressively, but for now (and probably even then), you're gonna really need to crank your stems to get things audible over the in-game sound effects.

There's a certain sweet spot of loudness, I've found, where things feel really HMX-like, and any quieter crawls up my ass to no end. I have rebuilt discs multiple times just trying to get things at the right volume. To minimize the amount of builds you might need to do (if you can even be assed), here's some rough tips for what to shoot for:

Limit, not amplify (sorta)

We measure the loudness of audio in decibels, usually. Digital audio has a hard ceiling of 0db. Any louder, or at that volume, and it distorts, and in a really unpleasant way (enjoy crackling!). Thus, you can't really just turn the gain on your stems up and hope it fixes your issue.

So what do you use instead? It's called a compressor, and its rather brutish cousin, a limiter. Compressors take the loudest points in a piece of audio and squash them down to give you a more even sound. It also lets you turn a track up much louder, as the loudness that we perceive isn't in the "spikes" (or peaks) in a waveform, it's in the flatter, broader "body" of the waveform.

Now, normally I hate compressors. If you've ever heard the term "loudness war", you can thank modern digital compressors. Not to mention, I've never found a mix that wasn't recorded like actual ass that has ever benefitted from compression. I normally need to use a compressor on, say, bass tracks that are so weak that boosting them without swamping the mix in low end isn't feasible.

However, for GH2, you'll need to compress by a lot to get it good and audible, and really, you're not gonna notice the headroom being gone anyway. Worse fates. If you've got a brickwall limiter you like, get it out. If not, the Limiter effect in Audacity is good enough (it's what I use for all my customs).

It's about average volume, not peak volume

There's two metrics by which to measure the loudness of a piece of audio, the peak and the average value. Peak is the absolute loudest the track gets. An errant click in a needle drop can cause a bad peak, but that's not representative of the rest of the audio. Average (or RMS) loudness refers to the rough overall level of the audio. A song might peak at 0db, but can have its average volume around -18db (which would be a pretty dynamic track).

Audacity by default doesn't let you RMS normalize, unfortunately, only "Amplify" and "Normalize". Amplify is a pretty crude boost. You say boost by 2db and it will, and if you check the "Allow clipping" box, it'll boost past the distortion point. Normalize is slightly more refined, setting the peak loudness of the audio to whatever you set the effect to use. But it's still only peak loudness. Thankfully, you can get an RMS normalization plugin that'll let you set average loudness instead.

My process

  1. Mix the track to three stems. Get everything at a nice volume, guitars, drums, effects, all of it. Not necessarily equal volume (though guitar should be right up there, since it's one of the two playable parts), but good levels. You should also be able to clearly hear the attack of the bass strings through the din. RBN/C3 docs recommend a cut under 60hz and low-mid boost around 600-1000hz generally. If you only have mixed audio (say, an MP3), ignore this step.
  2. Set the average loudness of the band track (or your MP3) to -13db. This is fucking loud, and you'll see a ton of peaking for most songs. This is alright, because it's not staying like that.
  3. Boost the guitar and bass stems, if applicable. Take note of how many decibels you go over by selecting the band stem in full, going into the Amplify effect, and noting how much it wants to decrease the signal by. This tends to be around -7db. Boost the guitar and bass stems up by that amount, or perhaps a few decibels under if they start to overtake the mix. If you've only got an MP3, skip.
  4. Soft limit every track to -0.5db. This will take off the peaking parts of each stem, avoiding any artifacting that'd come with a hard limit. You need to give the track at least half a decibel of headroom to avoid any audible clipping in-game.

Now, sometimes, if you have a drum track that's rather thin, doing this will cause you to outright lose your snare. The way to fix this is to limit in "stages", more gently limiting each of the drum stems, then gently limiting the full drums mixdown stem, then limiting the band track after that. This will, for lack of a better term, "fatten" the drum track and allow you to keep more of it audible on that final limit.

After your stuff's mixed, use RockAudio to convert your WAV to a VGS file.