How we invented On Stage Spatial Audio
THX certification exists. Apple Spatial Audio ships on every iPhone, every AirPod, every Mac. Automotive manufacturers are now negotiating speaker counts with luxury car buyers the same way they negotiate horsepower. The infrastructure for true spatial audio has been building for decades, quietly, in living rooms and cars and earbuds everywhere. And everyone is using it to watch TV.
The Standard Nobody Uses for Music
THX was invented to score cinematic immersion — how accurately a room reproduces the director's intent. The certification process is rigorous. Speaker placement, room acoustics, calibration, channel separation. All of it developed so that the experience of sitting in a theater could be replicated at home, precisely.
For film, this worked. The technology found its purpose.
For music, it found nothing. A mastered stereo track is a brick — every instrument compressed together into two channels, delivered flat. You can run it through a THX-certified 7.2.4 system and it will still sound like two speakers. Because it is two speakers. The certification is irrelevant when the source has already collapsed everything into a wall of sound.
Apple understood this and built Spatial Audio — head-tracked, object-based, genuinely impressive technology. What do the major labels give it to work with? Dolby Atmos upmixes. Algorithms guessing where the instruments probably were. Fabricated spatial information draped over a mix that was never designed for it. You can hear the seams.
Car manufacturers are putting multi-speaker arrays into vehicles. Meridian. Burmester. Bang & Olufsen. My car has 22 speakers. The music source handed to all 22 of them is a Spotify stream in two channels. Left and right. That's it.
The hardware kept getting built. The music never caught up.
A Tweet About StemDeck
I saw a tweet about an app called StemDeck — a DJ tool that separates a mixed track into individual instrument stems in real time. What caught my eye wasn't the app. It was the model behind it: Demucs. A hybrid transformer model trained to pull a mixed song apart into its constituent performances — vocals, guitar, bass, drums, piano, other. Not perfectly. Not without artifacts on everything. But well enough to be genuinely useful.
What Demucs had actually done was return the stems. The individual performances. The thing that gets locked away when a record is mixed and mastered and delivered to two channels.
Then I thought about my speaker system.
The mapping almost wrote itself. Vocals to the center channel — that's where the human voice belongs in any properly mixed spatial environment. Guitar is the widest stereo instrument in most rock recordings, so it goes front left and right. Drums belong in the rear — not because of theory, but because when you're standing in front of a band, the kick is behind you and slightly left, the snare is in front, the cymbals spread across the back of the stage. Bass is low and omnidirectional; it gets the left surround and the sub. Piano goes right surround — my surrounds sit further back than ideal, so I gave piano a bleed forward into the front right to compensate, and did the same for bass into the front left.
I wrote an ffmpeg filter graph that reads all six stems, routes them through an eight-channel mix, and outputs a 24-bit FLAC. Plex sees it as a standard multi-channel audio file. Infuse on the Apple TV plays it through the receiver with no configuration. The AVR handles bass management to both subs.
What It Actually Sounds Like
It doesn't sound like "surround sound music." That's a gimmick — a guitar pan that swings behind your head to prove the technology works. This doesn't announce itself at all. It just sounds like you're there.
The guitar comes from in front of you. The drums arrive from behind you and slightly left. The vocals are locked in the center, four feet in front of your face, the way they would be if you were watching someone sing. When the piano enters it comes from your right rear — where the piano would actually be on a stage that size.
It's not surround sound as a feature. It's accurate reproduction of where the instruments were.
"I Still Haven't Found What I'm Looking For" was an early test. Bono's voice in the center channel. Edge's guitar spread front stereo. The kick drum pushing through the LFE from behind you. That song was recorded at Red Rocks. The space is in the recording. Two speakers flatten it. Eight speakers let it breathe.
So Why Hadn't Anyone Done This
The stems were never available. Labels don't release them. A finished master is the product — the performances inside it are not yours to route anywhere.
Demucs changed that. Not perfectly. Highly compressed pop recordings artifact badly, and anything mixed intentionally narrow doesn't benefit much. But on live-room recordings, acoustic performances, anything where the original had space — the separation is good enough to work with. Good enough to build something real.
THX built the room. Apple built the earbuds. Car manufacturers built the arrays. Nobody pointed any of it at the stage.
We did.