CensorCraft

CensorCraft is a Forge Minecraft mod that actively listens to your voice with Simple Voice Chat and punishes you for saying forbidden words.

Demo It

Launch with Simple Voice Chat installed, hop into a single-player world, and say "boom" into your microphone.

How It Works

When you first try to join a world, you'll be asked to download a model. When in-game, your audio feed is transcribed using this model. If it detects you spoke a forbidden word, all sorts of punishment may be incurred, such as:

• Exploding the player
• Striking the player with lightning
• Summoning any mob by the player
• Giving the player any potion effect
• Teleporting the player to a random dimension
• Instantly killing the player
• Running any command
• … etc.

Punishments are set by the server admin. All punishment options are available in the server config file.

Models

A model is what CensorCraft uses internally to transcribe your voice. The models are hosted by HuggingFace here, and Minecraft servers can only request that you download a file from this location. Without a model, CensorCraft cannot operate.

Setup

In a typical setup, only server admins have to deal with configuring CensorCraft. Server players usually have a plug-and-play experience.

Server

world/serverconfig/censorcraft-server.toml is where server admins configure CensorCraft, such as changing the list of forbidden words, picking the appropriate punishment, etc. Most changes update live, so restarting the server isn't necessary.

Client

Players can use the config button found in the mods menu to change basic settings (hiding live transcriptions, changing latency, etc.):

Alternatively, players can edit /config/censorcraft-client.toml (the config screen and this file are synced).

Latency

Behind the scenes, CensorCraft breaks apart your live audio feed into individual recordings to transcribe your voice. Latency represents the delay (in milliseconds) between recordings. Lower latency means there are more recordings to process, and if your computer can keep up, you will get transcription results back faster. Raising the latency means you will get results back more slowly. Tinker with this value to get it just right for your machine.

Q/A

Q: When I first try to join a world, I'm asked to download a file. Do I have to, and is it safe?

A: Yes, and yes. Because models are too large to be bundled with the mod, CensorCraft downloads the model requested by the server to transcribe your voice before you join a world. All models are hosted safely by the HuggingFace community.

Q: I'm getting weird transcription results. What's happening?

A: Transcription accuracy is bound to the model the server requested. Although the server admin can increase transcription accuracy with a bigger and better model, transcription isn't always perfect. If you are getting particularly unusual results, delete the model from your client config folder and rejoin the world to install the model again. If it persists, open an issue.

Q: CensorCraft is far behind. Why is it so slow?

A: You may see a "CensorCraft is far behind" message if transcription is taking too long. This could be due to a number of factors, such as latency being set too low, or your hardware not being able to support transcription on the GPU (see performance). Try raising the latency in your client config menu to reduce the collection of backlogged audio samples.

Performance

By default, transcription is handled by the CPU. However, on Windows machines with GPUs that supports Vulkan, transcription is handled by the GPU by default to increase transcription speed (this behavior is toggleable).