Sixty-Four Kilobytes of Echo

The longest day of the year has become my excuse for a listening session. I wait until it finally gets dark, which in June takes forever, open the window, put on headphones and play old soundtracks until I’m too sleepy to choose the next one. This year I also fell down a rabbit hole about how that music was made, and I can’t stop thinking about the numbers.

A computer just for sound

Inside the Super Nintendo there is a second, separate little computer whose whole job is audio. It was designed at Sony: an 8-bit processor called the SPC700 paired with a digital signal processor that mixes eight voices at 32,000 samples per second. The main console sends it instructions, and it does the rest on its own.

It has 64 kilobytes of memory. Not 64 kilobytes for the music — 64 kilobytes for everything. The program that plays the song, the song data itself, every instrument sample and, as it turns out, the echo, all have to fit in a space a tiny fraction of the size of a single photo on your phone.

Squeezing the instruments

The instruments are short recorded samples, compressed with a scheme called BRR that packs sixteen samples into nine bytes. Most of them are tiny: a fraction of a second of flute or strings, looped so it can be held for as long as the note needs. When the chip plays a sample back at different pitches, it smooths between the sample points with a gentle interpolation that rounds off the high frequencies.

That smoothing is a big part of the sound people remember. It’s why so much of that music feels warm and a little muffled, as if you were hearing it through a wall in a house where someone you like is playing piano. Nobody chose it as a style. It’s arithmetic. But composers learned to write for it, the way painters learn what their pigments will do.

The price of echo

My favorite detail is the echo. The sound chip has a hardware echo with an eight-tap filter, which is how so many tracks from that era get their sense of space: the cave that rings, the cathedral, the long hallway at night. But the echo needs memory to remember what it just played, and that memory comes out of the same 64 kilobytes.

Every sixteen milliseconds of delay costs two kilobytes. The longest delay, 240 milliseconds, costs thirty. That is almost half of all the memory the sound system has. So every composer and sound programmer faced a very concrete trade: a deep, lush echo, or more instruments. A bigger hall, or a bigger orchestra.

Once you know this, you start hearing the budget. Some soundtracks are dry and dense, with many instruments packed in close. Others are sparse and wet, a few voices drifting in a lot of space. Neither is better. They’re different answers to the same question.

Snapshots of a machine

Fans preserved this music in a lovely way. An SPC file is a snapshot of that little computer at the moment a track begins: its entire 64 kilobytes of memory plus the state of the sound chip, about 66 kilobytes in all. A player that emulates the chip simply switches the snapshot back on, and the song plays exactly as the hardware would play it. A whole soundtrack fits in the space of a couple of ordinary MP3s.

I like that the music isn’t stored as audio at all. It’s stored as the machine, paused.

People who wrote inside the box

Knowing the limits makes me admire the people who worked inside them even more. Yasunori Mitsuda wrote most of Chrono Trigger’s score as his first big job and pushed himself so hard that he ended up with stomach ulcers; Nobuo Uematsu stepped in to help finish it. Uematsu’s own score for Final Fantasy VI still sounds far bigger than the hardware should allow. Hiroki Kikuta’s music for Secret of Mana somehow makes eight voices sound like a forest full of birds. And back in 1990, Yuzo Koshiro’s music for ActRaiser made a lot of people realize what this little chip could do.

This year’s longest-day playlist

“Corridors of Time” — Chrono Trigger, Yasunori Mitsuda
“Fear of the Heavens” — Secret of Mana, Hiroki Kikuta
“Beware the Forest’s Mushrooms” — Super Mario RPG, Yoko Shimomura
“Terra” — Final Fantasy VI, Nobuo Uematsu
“Onett” — EarthBound, Keiichi Suzuki and Hirokazu Tanaka

I played them in that order on closed-back headphones, with the window open and the city slowly going quiet. Somewhere around the third track the streetlights came on. Somewhere around the fifth I stopped analyzing anything and just listened, which, after all the numbers, is still the best way to hear it.