CensorCraft

The mod works with Simple Voice Chat. While a player's microphone is unmuted, the in-game audio feed is transcribed by a speech model, and any detected forbidden word triggers the punishment that the server owner has configured. Punishments can include an explosion, a lightning strike, a summoned mob, a potion effect, a teleport to a random dimension, an instant kill or a command. Other punishments can be added through CensorCraft plugins.

Because the speech model is too large to bundle with the mod, CensorCraft fetches it into the config folder on the first launch of a world, and the models are hosted on HuggingFace. Transcription runs on the GPU when Vulkan is available and falls back to the CPU otherwise, which is slower. Players who see long delays in messages can install Vulkan, update graphics drivers, or turn off the Use GPU option in the client config to use the CPU instead. The mod runs on client and server, so both sides need it, and it is released under the MIT License as open source.