Audio Programming

On a home computer of the 1980s, sound was a list of numbers. A program played a note by writing a pitch, a volume and perhaps a waveform into the registers of a sound chip, and the chip kept making that sound until the program wrote something else. There was no library to call and no operating system to mix voices. A tune was a small program, usually run fifty or sixty times a second, that kept rewriting those registers; a sound effect was the same thing, interrupting the tune. And on the cheapest machines there was no chip at all, only a speaker the processor had to switch on and off for every cycle of every note.

Every machine drew the lines in a different place. The same tune had to be rethought for three square waves on one machine, three synthesiser voices on another and four channels of recorded sound on a third, and the people who did it well became known by name.

Fast facts

• The method: write values to the sound hardware’s registers; nothing else stands in between.
• The timing: usually a music routine called from an interrupt fifty or sixty times a second (see Interrupt-Driven Music); beeper music on the 48K Spectrum had to count out every cycle instead.
• The software: a sound driver that reads the composer’s music data and feeds the chip.
• The tricks: arpeggios for chords, software envelopes and vibrato, pulse-width modulation, filter sweeps, sampled sound.
• The budget: a few kilobytes of memory and a slice of every frame’s processor time.

What each machine offered

The clock behind the chip mattered as much as the chip. The Spectrum 128 runs its AY-3-8912 at 1.7734 MHz and the Amstrad CPC runs the same part at 1.00 MHz, so the same period value plays two different notes. A C64 note table written for an American machine plays flat on a European one.

On the Commodore 64

Commodore’s own manual made no secret that the programmer was on their own. The Programmer’s Reference Guide introduces the SID as “one of the most sophisticated electronic music synthesizers available on any computer”, with “three voices, totally addressable, ATTACK/DECAY/SUSTAIN/RELEASE (ADSR), filtering, modulation, and ‘white noise’”, and then explains how to reach it: “You get to the sound generator by POKEing into specified memory locations.” The 25 registers a program writes run from 54272 ($D400) to 54296. “The workhorse of your sound programs is the POKE statement.”

This program plays one note, the A above middle C, on voice 1:

10 S=54272
20 FOR R=S TO S+24:POKE R,0:NEXT
30 POKE S+24,15
40 POKE S+5,0:POKE S+6,240
50 POKE S+1,29:POKE S,69
60 POKE S+4,17
70 FOR T=1 TO 2000:NEXT
80 POKE S+4,16

Line 20 clears every register. Line 30 sets the master volume to 15. Line 40 gives voice 1 an instant attack and decay and a full sustain level. Line 50 writes the 16-bit frequency value 7493 ($1D45) as a high byte, 29, and a low byte, 69. Line 60 selects the triangle wave and sets the gate bit, which starts the note; line 80 clears the gate, which starts the release. The data sheet’s formula turns the frequency value into a pitch, value × clock ÷ 16,777,216, so the right value depends on the machine’s clock: 7493 is A at 440 Hz on a European (PAL) C64, as the SID chip entry’s frequency tables show. The Programmer’s Reference Guide’s own value for that note is 7217, which plays at about 424 Hz on a PAL machine. Commodore’s table was worked out for the faster American NTSC clock, so a European program that copied it played about two-thirds of a semitone flat.

The music was also a memory budget, and on a commercial game the composer’s share was argued over. Chris Butler, who wrote Elite’s C64 conversion of Commando in about eight weeks for Christmas 1985, told ZZAP!64 what happened to its third level: “the game was completely finished — but when Rob Hubbard came down he said that he needed about 6K for the music, and there wasn’t room. So Steve Wilcox, the man, said, ‘Right — hack out level three.’” Butler, “not a musical person at all”, still rated the machine: “considering it’s probably got the best music capabilities of any home computer, excluding the 16 bit stuff… it’s not bad at all.”

The music routine

Most music on these machines came from the same shape of program, called from an interrupt once a frame. Trace this:

each interrupt (50 times a second on a PAL machine):
    count down to the next row of the tune
    when a row is due: read the next notes from the music data
    for each channel:
        update the volume envelope
        update the pitch (vibrato, slides, arpeggio)
        write the new values to the chip's registers
    let any sound effect overwrite the channel it has taken

Each pass is a “tick”, and the tempo is a count of ticks per row of the tune. The trackers made the arithmetic visible. In the documentation of the ProTracker module format, the default is 6 ticks a row and 125 beats a minute, with four rows to a beat: at 50 ticks a second that is a row about every eighth of a second.

The 48K Spectrum is the exception that shows why the interrupt mattered. Its ROM’s BEEP routine switches interrupts off while it plays, and the processor does nothing else until the note ends. A game with continuous music had to slot the notes in between everything else. Matthew Smith told Sinclair User in 1984 that the music in Jet Set Willy “does not use an interrupt”: “It’s just move-a-tiny-little-bit, BEEP-a-tiny-little-bit.” Beeper Music tells that story.

Tricks the hardware did not provide

Programmers filled the gaps in the hardware with software, and many of the best-known sounds of the period are those workarounds:

Some of these cost a great deal. Sega’s own manual warned Mega Drive programmers that with the sample DAC “all timing must be done by software”, and that unless the program was very carefully built it could not use the FM channels at the same time. The SID’s sample trick took a large share of the C64’s processor time too. On the Amiga, OctaMED, which Amiga Format described in 1992 as “the 8-channel version of MED”, could only play eight voices on Paula’s four channels by mixing them in software.

The AY’s envelope trick comes straight from the data sheet. General Instrument’s manual explains that the envelope’s frequency is the input clock divided by 256 and then by the 16-bit envelope period, and that the generator then steps through “a 16-state per cycle envelope pattern”. With a repeating shape, a sawtooth or a triangle, and a small period, that cycle falls in the audible range, and the envelope becomes a buzzing waveform the tone channels cannot make.

Writing the music down

Composers worked in three main ways.

Straight into the program. Early composers wrote their notes as data for a routine they had written themselves. Rob Hubbard started that way: after hearing Activision’s Master of the Lamps he connected his monitor to his hi-fi and wrote his own music utility, because he found the commercial ones “very limited with the sounds you can create” (ZZAP!64, 1985).

Music Macro Language. Some BASICs played tunes written as text. MSX BASIC’s PLAY statement “plays music according to the music macro language”, taking up to three strings, one for each of the AY’s voices: letters A to G for notes, O for the octave, L for the note length and T for the tempo, with M and S setting the period and shape of the chip’s hardware envelope. PLAY "T120O4L4CDEFGABO5C" plays a C major scale. Microsoft’s GW-BASIC for the IBM PC has a PLAY statement built on the same idea.

The tracker. On the Amiga, music was written in a tracker, a grid of notes, sample numbers and effect commands that scrolls up the screen, one column per channel. Karsten Obarski’s SoundTracker began it at the end of 1987, and ProTracker became the standard; its files are the MOD format. The same way of working later spread to other machines, including newer trackers for older chips, such as SID-Wizard for the C64 and FamiTracker for the NES.

See also

• SID Chip
• AY-3-8910
• Paula
• APU
• YM2612
• FM Synthesis
• Sound Drivers
• Interrupt-Driven Music
• Beeper Music
• Hard Restart