Iron's Spells: Incantation

Say a spell's name. Cast it.

Incantation captures your microphone directly, runs the audio through an offline speech recognizer (Vosk), and casts whichever Iron's Spells 'n Spellbooks spell you just said. No voice training, and the speech model installs itself on first launch.

How it works

Incantation opens your microphone itself — no voice-chat mod involved — and feeds the audio straight to a Vosk recognizer whose grammar is built from the spell names your installed mods actually register. A closed grammar's only way to say no is its [unk] token, and it will often prefer the least-bad spell name over it - so 0.10.6 adds the defences that make that safe: a match has to have been spoken long enough to plausibly be that phrase, and any spell you list in perSpellMinConfidence waits for the complete utterance and its own confidence bar before it will fire.

Casting is then gated on what you actually have equipped: a spellbook in your Curios slot, an imbued weapon you're holding, or imbued armor you're wearing. Saying the name of a spell you don't have does nothing. Servers that would rather let you cast from a spellbook held in hand can set castMode = ANY_SPELLBOOK.

Recognition runs entirely on your machine. No accounts, no telemetry, and your microphone audio is never transmitted, never recorded, never written to disk. The one time Incantation touches the network is the speech-model download on first launch; set autoDownloadModel = false and install the model yourself if you would rather it didn't.

When it listens

By default the microphone is only open while you have a spellbook, staff, imbued weapon or imbued armor — in either hand, worn, or in your Curios slot. Stow or remove it and the mic closes. No keybind involved.

That means conversation with your spellbook stowed can never cast anything, which matters because a lot of people run a voice-chat mod on the same microphone. But if you keep a spellbook in your Curios slot permanently, this is close to ALWAYS_ON in practice — so if you talk to friends mid-fight, use HOLD_KEY_AND_ITEM. Carrying a spellbook isn't on its own evidence you meant to cast; that mode requires a held key too, so speech never reaches the recognizer unless you asked.

Set gatingMode to taste:

These gate capture, not recognition: in every mode but ALWAYS_ON the recognizer never receives the audio, and the device is closed rather than merely ignored. The mic is also released whenever the game is unfocused or paused.

The HUD's mic chip tells you which state you're in at a glance — the dot is dim when the mic is closed, lit when it's open, and pulses while it's actually hearing you, with a live level meter beside it.

Features

• Speak any indexed spell — works with Iron's Spells, all its addons, and any mod that registers spells through Iron's API.

• Equipped-only casting (default ON) — a spell only fires if you actually have it on you. Anything else is rejected before it leaves your client.

• Imbued weapons and armor supported — hold an imbued sword or staff, or wear an imbued chestplate, say the spell it carries, it casts via the same mechanism as right-clicking.

• HUD chip — corner-anchored chips showing a live mic indicator and level meter, plus last cast, recent cast history, queued spells, miss toast, the last phrase heard, and "did you mean…?" alias suggestions.

• Vanilla advancements — voice-cast milestones (1, 10, 50, 200, 1000) and combo casts surface as standard advancement toasts.

• Spell List screen — browse every spell the registry exposed with the exact phrase Vosk listens for, plus per-spell cast counts and school filtering.

• Voice Codex — personal stats (top-cast spell, daily streak, median latency).

• Test Arena — safe practice mode that records what would have cast without firing anything. Useful for tuning aliases or learning a tricky spell's pronunciation.

• Welcome wizard on first launch — walks you through mic + model + first cast.

• Aliases & custom phrases — bind words the model can pronounce to any spell id.

• Play in your own language — Spanish, Russian and French speech models are one config line away (modelId), and config/voicespells/phrasebook.json is generated with every installed spell and an editable override field, so you can translate the whole spell list in one file instead of adding aliases one at a time. Your overrides survive updates, and newly installed spells are appended automatically.

  The model is only half of it. The recogniser only listens for words its model knows, so the spell phrases have to be translated too — a Russian model against English spell names hears nothing at all. The repository's contrib/ folder has a Spanish phrasebook contributed by the community, translated and ready to use. The Russian one is a blank template: 294 entries with every override empty, because the machine-translated version that used to fill it had never been spoken into a microphone, and a wrong phrase gets trusted where a blank one asks to be filled. Russian recognition therefore needs that file filled in first. /voicespells vocab tells you which of your phrases the loaded model can actually pronounce.

  German does not work yet, and not for want of a model: Iron's Spells ships no German translation, so there are no German spell names to start from. It needs a hand-written phrase pack. If you speak German and fancy writing one, please get in touch.

• Loadout shortcuts — say one word, cast the first castable spell from a list (cooldown + mana aware).

• Voice commands — say no to clear spells waiting in the cast queue, or yes to accept a "did you mean…?" alias suggestion (enable handsFreeConfirm first).

• Reward speaking (server, off by default) — voice casts can come back faster, hit harder, or both: voiceCooldownPercent scales the cooldown, voiceLevelBonus adds spell levels, and the extra levels are charged at the spell's ordinary mana cost rather than the higher one. Set them per player if you want, with playerAdvantages:

  playerAdvantages = [
      "Steve=cooldown:50,level:2",   # Steve casts faster and harder
      "!Alex",                       # Alex plays by ordinary Iron's Spells rules
  ]

  Out of the box nothing is changed — the defaults are plain Iron's Spells. Turn on voiceVolumeScaling and the bonus has to be earned: the level your item itself would cast at — the spellbook, the imbued item, or the scroll, affinity bonus included (level 1 under FREE) — is always the floor, and how much of voiceLevelBonus a cast adds on top depends on how loudly you said it: a whisper earns none, your normal speaking voice about half, a raised voice all of it — rounded to a whole level, judged against your own calibrated speaking voice.

• Incantation-only mode (server, off by default) — require spells to be spoken. FIRST_CAST makes a player voice-cast a spell once before they can click it, so learning the incantation unlocks it; ALWAYS means spells can only ever be spoken.

• Server-side controls — per-player whitelist, blocklist, rate limit, broadcast-nearby, cast logging.

Requirements

• Minecraft 1.21.1 (NeoForge 21.1.x) or 1.20.1 (Forge 47.x)
• Iron's Spells 'n Spellbooks by iron431
• Curios API — recommended, and an Iron's Spells dependency anyway
• A microphone
• No voice-chat mod required. Incantation captures the microphone itself, and coexists with Simple Voice Chat if you use one to talk to friends.

Setup

1. Drop the Incantation jar in your mods/ folder alongside Iron's Spells.
2. Launch the game. On first run Incantation downloads vosk-model-small-en-us-0.15 (~40 MB) into config/voicespells/ for you.
3. Join a world, equip a spellbook (hand or Curios slot), and say a spell it carries.

Prefer to install the model yourself? Set autoDownloadModel = false, grab one from alphacephei.com/vosk/models, and unzip its contents into config/voicespells/model/ so you end up with config/voicespells/model/am/, config/voicespells/model/conf/, etc.

For tricky spell names (Traveloptics, Cataclysm, anything with unusual phonetics), map an alias in config/voicespells-client.toml:

customPhrases = [ "dark beam=traveloptics:abyssal_blast" ]

then use More… → Reload grammar now — no restart needed.

Configuration

Most settings are accessible in-game from Mods → Iron's Spells: Incantation → Config.

• Recognition tab — gating mode, equipped-only restriction, fuzzy tolerance, substring match, dedup window, debug monitor.
• HUD tab — corner, offset, opacity.
• More menu — welcome wizard, Voice Codex, Diagnostics, Test Arena, reload grammar, profile export/import, calibrate mic.

Client commands: /voicespells devices lists every capture device the game can see and which one is selected — start there if the microphone looks dead.

Server settings (config/voicespells-server.toml): cast mode (CURIO_SPELLBOOK / ANY_SPELLBOOK / FREE), per-player whitelist, blocklist, rate limit, broadcast-nearby radius.

Troubleshooting

• Nothing happens when you speak — check the gating mode first. By default the mic is only open while a spellbook, staff, imbued weapon or imbued armor is in your hands, worn, or in your Curios slot. The HUD mic dot tells you: dim means closed.
• Chat says "Vosk model not found" (or Diagnostics shows Vosk model: FAIL) — no model under config/voicespells/model/. Recheck the layout: am/, conf/, graph/ should sit directly under that path.
• Microphone looks dead — run /voicespells devices to see what the game can find, and set captureDevice if the default isn't the one you want.
• Recognition feels slow — open More → Calibrate mic and say a few spell names. Or use the Diagnostics screen to verify each prerequisite.
• Test Arena doesn't cast — by design. It records would-be casts so you can practice safely.

License

MIT.

Credits

• Vosk — Alpha Cephei (Apache 2.0)
• Iron's Spells 'n Spellbooks — iron431
• Spanish phrasebook — NeoTargetStudios (Spidercat0926), who translated close to 300 spell names across nine spell mods and helped with the mod itself