🧙 VoiceMage

VoiceMage is a Fabric mod for Minecraft 26.2 that turns your microphone into a spellbook. Say "fireball" and a fireball flies. Say "meteor" and the sky answers. Speech recognition runs locally on your own machine, so there is no cloud, no account and no audio ever leaves your computer. The server decides what actually happens: mana, cooldowns and spell effects.

Everything is tweakable in voicemage.json or live with commands: spell costs, mana regeneration, whether the Explosion needs its legendary chant, and a lot more.

✨ Features

🎙️ Voice Casting

• 🗣️ 14 Spoken Spells: Say the name of a spell and it fires in the direction you are looking.
• 📴 Offline Recognition: Powered by Vosk. The speech model (~40 MB) is downloaded once and then works fully offline.
• 🎤 Microphone Picker: Choose any recording device from a list with /voicemic. Devices that can not record at 16 kHz mono (many USB headsets and laptop microphones) are converted automatically.
• ⚡ Instant Mute: Press V to mute or unmute. The speech model stays in memory, so there is no waiting.
• 🧩 Smart Spelling: Common variants such as "fire ball", "pit fall", "lightening", "magic missle" and "poly morph" are understood too.

🔮 Mana & Combat

• 💧 Mana System: A mana pool with an on-screen bar, configurable size and regeneration.
• ⏱️ Cast Cooldown: A short delay between casts keeps spamming in check.
• 🎯 Aimed Spells: Lightning, Magic Missile, Meteor, Polymorph, Shrink and Explosion go where you look.
• 🛑 No Wasted Mana: A spell that has nothing to do (Heal at full health, Pitfall with no blocks, Shrink without a target) costs nothing.

🛡️ Built for Servers

• 🖥️ Server Decides: Mana, cooldowns, range checks and every spell effect are handled on the server.
• 🛠️ Live Admin Commands: Change mana costs, regeneration and the Explosion chant without restarting.
• 💾 Safe Config: An unreadable voicemage.json is backed up instead of overwritten, and /voicemage reload keeps your old settings when the file is broken.
• 🧹 Self-Cleaning: Temporary walls, ice cages, clones and polymorph animals are removed even after a crash.

🪄 The Spells

Costs are the defaults and can be changed with commands (see below).

💥 The Explosion Chant

Explosion costs a full mana bar and, by default, has to be chanted. Either of these works:

O ancient flame, awaken from your endless slumber. Let the skies burn and turn my enemies to ash, leaving nothing behind. EXPLOSION!

Darkness blacker than black and darker than dark, I beseech thee, combine with my deep crimson. Return all creation to cinders, and come from the abyss! EXPLOSION!

Server operators can allow casting by saying just "Explosion" with /voicemage chant false.

⚠️ Heads up for server owners: Explosion carves a very large crater (radius 38, depth 128 by default) and Pitfall removes blocks down to the bottom of the world. Both can seriously damage builds. Raise the costs, keep the chant required, or tune the crater size in the config before you let players loose.

🎮 Commands & Permissions

Server commands

Commands marked Op need permission level 2 (game master), such as operators and command blocks.

<spell> is one of: fireball, pitfall, heal, lightning, earth_wall, blink, magic_missile, meteor, wind_blast, freeze, clone, polymorph, shrink, explosion.

/voicemage regen 10          # regenerate 10 mana per second
/voicemage fireball 20       # make Fireball cost 20 mana
/voicemage chant false       # just say "Explosion" to cast it
/voicemage reload            # apply manual edits to voicemage.json

Client commands (only affect you)

Your choice is saved in config/voicemage-client.json. If the saved device is not connected, the system default is used.

Keybind

⚙️ Installation

1. Install Fabric Loader 0.19.3 or newer for Minecraft 26.2, and the Fabric API.
2. Make sure you run Java 25 or newer.
3. Drop voicemage-26.2-<version>.jar into the mods folder of both the server and every player.
4. Start the game and join a world or server.
5. First launch only: VoiceMage downloads the English speech model (vosk-model-small-en-us-0.15, about 40 MB) into <game directory>/voicemage/. The download is checked against a built-in size and SHA-256 checksum and thrown away if it does not match.
6. Allow microphone access if your operating system asks. When the HUD shows Voice: listening, you are ready.

Players without the mod can not cast, and the server needs the mod to run the spells.

🧠 How It Works

Every spell goes through the same three steps:

1. Listen: Your own client hears you. Vosk only knows the spell names and the chants, so random talk is mostly ignored.
2. Ask: The client tells the server which spell you want. No audio is ever sent.
3. Decide: The server checks that you are alive, off cooldown and have the mana, then runs the effect and takes the mana. If the spell can not work (no target, nothing to heal), it costs nothing.

The HUD shows your mana bar and a voice status:

🧩 Configuration

The config lives in config/voicemage.json and is created on first start. Edit it and run /voicemage reload, or change the main values in-game with the commands above. Values outside the allowed range are clamped automatically.

{
  "maximumMana": 100.0,
  "manaRegenerationPerSecond": 5.0,
  "minimumCastIntervalTicks": 6,
  "explosionChantRequired": true,
  "pitfallEnabled": true,
  "fireballManaCost": 35
}

If the file can not be read (for example after a typo), VoiceMage keeps a copy as voicemage.json.broken-<date>-<time> next to it instead of silently overwriting your settings.

🛠️ Compatibility

• Minecraft Version: 26.2
• Mod Loader: Fabric (Loader 0.19.3+) with the Fabric API
• Java Version: Java 25+
• Environment: Client and server (voice runs on the client, spells run on the server)
• Languages: The speech model is English only

Good to know

• The server can not hear your microphone. It decides mana, cooldowns and effects, but whether the Explosion chant was spoken is checked by the player's own client.
• The microphone needs a working input device that Java can use. On Linux, make sure PulseAudio or PipeWire exposes an input device.
• Temporary things are cleaned up even after a crash: leftover clones and polymorph animals are removed when their chunk loads, and temporary walls and ice are tracked in voicemage_temporary_blocks.json in the world folder and removed on the next start.

🩺 Troubleshooting

• "Voice error: Microphone ... cannot be opened": run /voicemic and try another device. The log (latest.log) says which device and which conversion is in use.
• The speech model does not download: the first launch needs internet access to alphacephei.com. Check your firewall or restart the game (a failed download is retried once automatically). If the log reports a checksum failure, the file was corrupted or changed and was discarded.
• Spells are not recognized: speak clearly and say one spell at a time. The recognizer is English only.
• Nothing happens on a server: the mod must be installed on the server too, and your mana must cover the spell's cost. Use /spells to see the current costs.

🏗️ Building From Source

You need JDK 25.

git clone https://github.com/Berkayprof/VoiceMage.git
cd VoiceMage
./gradlew build

The mod jar is written to build/libs/voicemage-26.2-<version>.jar. To test in a development client:

./gradlew runClient

src/main/java/com/voicemage/
├── Voicemage.java            server/common entrypoint
├── VoicemageClient.java      client entrypoint (voice, HUD, keybind)
├── VoicemageCommands.java    /spells and /voicemage
├── VoicemageNetworking.java  packet registration
├── client/                   speech recognition, microphone, model installer, HUD and visuals
├── config/                   voicemage.json handling
├── mixin/                    Mannequin accessor (used by Clone)
├── network/                  payloads between client and server
└── spell/                    spells, mana, effects, temporary-block cleanup

📜 Credits & License

• VoiceMage was created by Cayce and is licensed under the MIT License. The copyright notice stays with the original author.
• This version was rebuilt from the compiled mod into a Gradle project and is maintained by Berkayprof.
• Speech recognition uses Vosk (Apache License 2.0), bundled with the mod.

Original mod by Cayce · Maintained with ❤️ by Berkayprof