Tones, Noise & Pitch#

Two registers shape what a channel sounds like: AUDC picks the waveform, and AUDF picks the pitch. Together they’re the source of both the VCS’s distinctive voice and its most infamous flaw — it can’t play in tune.

AUDC: pick a voice#

AUDC is a 4-bit selector (0–15) that routes the channel through different internal divider and polynomial-counter combinations. The result isn’t 16 evenly-graded timbres but a handful of distinct families — pure tones, buzzy/reedy tones, and noise. The ones you’ll reach for most:

The exact character of all sixteen is genuinely best learned by ear in Stella — sweep AUDC through its values with a fixed AUDF and listen. But two anchors carry most games: 4 for a melody tone and 8 for noise.

The Stella Programmer’s Guide lists what each of the sixteen AUDC values does — though it, too, recommends experimenting to find your sound.

AUDF: pick a pitch — sort of#

AUDF is a 5-bit divisor, 0–31. The channel’s frequency is the fixed audio clock divided by AUDF + 1 (and by the waveform’s own divider). Two consequences fall out of that:

• Only 32 steps. Each waveform can produce just 32 base pitches. That’s it.
• The steps are wildly uneven. Because frequency is clock ÷ (AUDF + 1), the available pitches follow a harmonic series, not a musical scale. Near the top (AUDF = 0, 1, 2…) consecutive steps are nearly an octave apart — you simply cannot get most notes up there. Near the bottom (AUDF = 29, 30, 31) the steps are a few cents apart — far finer than you need. All the resolution is in the bass; the treble is a coarse ladder.

Musical notes, by contrast, are spaced in equal ratios — every semitone is the same multiplier. Those ratios almost never line up with the integer divisors, so most notes land audibly sharp or flat. This is why VCS music has its unmistakable slightly-wrong charm: the hardware physically can’t hit a tempered scale.

Living with the out-of-tune problem#

You don’t compute pitches at run time; you build a note → AUDF lookup table, choosing for each musical note the divisor (and sometimes the AUDC mode) whose frequency lands closest:

NoteTable:          ; AUDF value for each note your song uses
    .byte 31        ; a low note — lands close, lots of resolution here
    .byte 23
    .byte 16
    ; ...

A song then refers to notes by index into this table — which is exactly what the Song data in xmas/xmas.asm does, and what the next page plays back. Composers either lean into the detuning as part of the sound, or write melodies that favor the notes the hardware happens to hit cleanly.

In Practice#

• Build the note table by ear, not by formula. A computed equal-tempered table will be uniformly a little off; hand-picking the nearest divisor per note (and auditioning it) usually sounds better than the math.
• Switching AUDC can rescue a note. A pitch that’s badly off in one waveform’s range may land closer in another’s. Bass notes especially benefit from the lower-base modes (12/13).
• Noise has “pitch” too. AUDF still applies under AUDC 8 — a low AUDF gives a deep rumble, a high one a thin hiss. An explosion is often noise with AUDF ramped downward over a few frames.