Iron's Spells: Incantation

Say a spell's name. Cast it.

Incantation listens to your Simple Voice Chat mic, runs the audio through an offline speech recognizer (Vosk), and casts whichever Iron's Spells 'n Spellbooks spell you just said. No voice training, no extra keybind, and the speech model installs itself on first launch.

How it works

When Simple Voice Chat is transmitting your audio, Incantation feeds the frames to a Vosk recognizer built from the names of every spell in the registry. Whatever it hears is then checked against what you actually have equipped - a spellbook in your Curios slot, or an imbued weapon you're holding - and anything else is rejected on your client before it ever reaches the server. Servers that would rather let you cast from a spellbook held in hand can set castMode = ANY_SPELLBOOK.

The grammar deliberately stays broad. Narrowing it to just your equipped spells sounds safer, but with only two or three phrases left the recognizer force-matches any audio - a cough, a footstep, someone else talking - into whichever of them is closest. Keeping every spell name in there dilutes that, and the equipped check is what actually stops an unowned spell from casting.

Recognition runs entirely on your machine. No accounts, no telemetry, and your voice never leaves your PC - it 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.

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 supported - hold an imbued sword or staff, say the spell it carries, it casts via the same mechanism as right-clicking.
• HUD chip showing last cast, recent cast history, queued spells, miss toast, the last phrase heard, and "did you mean...?" alias suggestions.
• Themes - accent presets (Arcane, Blossom, Ocean, Dusk, Mint, Gold, plus Phoenix / Frost / Verdant / Necrotic unlocked by cast milestones) and base palettes (Dark, Midnight, Slate).
• Vanilla advancements for voice-cast milestones and combo casts.
• Spell List screen browsing every indexed spell with its exact phrase.
• Voice Codex - personal stats (top-cast spell, daily streak, median latency).
• Test Arena - safe practice mode.
• Welcome wizard on first launch.
• Aliases & custom phrases - bind words the model can pronounce to any spell id.
• Loadout shortcuts - say one word, cast the first castable spell from a list.
• Quick recast - one keybind (]) repeats your last voice-cast spell, for when saying the phrase again is slower than the cooldown.
• Voice commands - spell one through spell nine for slot select; yes/no for queue control.
• Play in your own language - 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. Speaking a language other than English also needs a matching Vosk model - Vosk can only hear words in the loaded model's lexicon.
• Server-side controls - whitelist, blocklist, rate limit, broadcast-nearby, cast logging.

Requirements

• Minecraft 1.21.1
• NeoForge 21.1.x
• Simple Voice Chat (henkelmax, 2.5.x+)
• Iron's Spells 'n Spellbooks (iron431, 3.x+)
• A Vosk English model - downloaded for you on first launch (~128 MB)
• Curios API recommended

Setup

1. Drop the Incantation jar in your mods/ folder alongside SVC and Iron's Spells.
2. Launch the game. On first run Incantation downloads vosk-model-en-us-0.22-lgraph (~128 MB) into config/voicespells/model/ for you.
3. Join a world, trigger SVC, and say a spell you have equipped.

Prefer to install the model yourself? Set autoDownloadModel = false, grab one from https://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. vosk-model-small-en-us-0.15 (~40 MB) is enough for vanilla spell names.

For tricky spell names, map an alias in config/voicespells-client.toml (or edit phrasebook.json) and use More... -> Reload grammar now - no restart needed.

License

MIT.

Credits

• Vosk - Alpha Cephei (Apache 2.0)
• Simple Voice Chat API - Maximilian Henkel (MIT)
• Iron's Spells 'n Spellbooks - iron431