Animalese: the text → speech pipeline

How Animal Crossing (GameCube, GAFE01, USA rev 0) turns a message string into villager babble. Every claim here is traced to a function in the ac-decomp tree.

The short version

message text  (game charset, 1 byte/char)
  │
  │  1. mMsg_sound_voice_get()            256-entry table
  ▼     char code ──► voice code (0x53–0x86)
gating + voice selection
  │  2. mMsg_sound_voice_mode_get()       animalese / click / silent
  │     mNpc_GetNpcSoundSpec()            speaker ──► spec 2–9
  ▼     Na_SpecChange()                   spec ──► sequence 243/244/245
windowing
  │  3. mMsg_sound_voice_entry()          collect (a, b, c) over two cursor steps
  ▼     ~1 character per 2 game frames
draw-time trigger
  │  4. mFontSentence_animal_voice_se()   glyph scale ──► volume
  ▼     sAdo_VoiceSe(a, b, c, npc, scale, mode)
phonology
  │  5. Na_VoiceSe()                      digraphs, long vowels, silent e
  ▼     ──► phoneme code 0x01–0x77
playback
  │  6. Sou_VoiceStart()                  write port 0 of a JAudio2 subtrack
  │     sequence dyntbl ──► instrument (code − 1)
  ▼  7. ProgToVp(bank, inst) ──► NoteToVoice() ──► smzwavetable
DSP  ──►  one 30–400 ms VADPCM sample, 22050 Hz, pitched down 4–9 semitones

A phoneme code is literally a soundfont instrument number plus one. The whole "engine" is a lookup table with about fifty special cases; the character comes almost entirely from the samples.

Pages

• 1. Gating & voice selection When animalese plays at all, and which of the three voices you hear. Specs 2–9, the click fallback, and the player's own voice.
• 2. Text → voice code The game's character set, the 256-entry voice table, case folding, accent folding, and the eight punctuation codes.
• 3. Windowing & timing How three characters get bundled into one call, why it fires from the font renderer, and the every-other-frame cursor.
• 4. The phonology engine Na_VoiceSe in full: the 53-entry digraph table, short/long vowel tables, silent e, magic e, and punctuation flush.
• 5. Playback Sequences, banks, the two alternating subtracks, per-spec timbre, the emotion LFO, and the deferred second syllable.
• 6. The soundfont lookup Sequences, banks and instruments as data: the four archive tables, the bank blob, load-time relocation, and where the PCM actually lives.
• 7. The editor path Mode 1 — an undocumented third calling convention that bypasses phonology entirely, and what it says about samples 0x53–0x76.
• 8. Walking the disc image CISO, the GameCube filesystem, the DOL, the bank blobs, the VADPCM decoder, and the voice sequences disassembled — every offset needed to extract and replay the samples from a dump.
• 9. Quirks, dead code & corrections The dead volume store, the tautological dedup, vestigial mouth-shape tracking, and corrections to earlier notes.
• 10. Reference tables Everything in one place: the full voice table, digraphs, phoneme tables, control codes, disc offsets, source line index.

Scope

Pages 1–7 document the runtime pipeline as decompiled. Page 8 documents the data: how to walk a dumped disc image down to individual VADPCM frames, with the audio table-of-contents, bank/sample formats, sequence disassembly and decoding notes. Together they are intended to be a complete reference for a reimplementation.

Verified against ac-decomp at commit 09ca8e8b (DOL and REL 100 % linked) and against a dumped GAFE01 USA rev 0 disc image.