VoiceSniffer / Overview

VoiceSniffer

Last modified: 25 July 2026

Get your server running

Follow these steps in order. Pick the processing path before you copy a configuration file.

Choose a processing path

VoiceSniffer has three engine modes. The entitlement rule is the part that catches people during setup:

Cloud

Use sidecar. The processor token is the entitlement, so cloud mode does not need a licence key in the licence configuration.

Self-host

Use sidecar with your Docker processor. Self-host mode requires the VoiceSniffer licence key and the processor token for this server.

Embedded

Use embedded when inference should run inside the Minecraft server JVM. Built-in mode requires the VoiceSniffer licence key and no Docker processor.

See engine modes for the exact settings, then use the config.yml reference to check every key.

What you need

  • Paper 1.21.11 and Java 21.
  • Simple Voice Chat API 2.6.20, installed and working on the same server.
  • Docker only when you run a self-hosted sidecar processor.

Cloud mode uses its processor token as the entitlement. Self-host and embedded modes require the licence key. Full details are on requirements.

What VoiceSniffer does

VoiceSniffer is in development and is not on sale yet. There is no public download. It launches first on BuiltByBit; release news goes out in the Sniffer Studio Discord. These pages describe the build as it exists today so that setup is not a surprise on launch day.

VoiceSniffer moderates Simple Voice Chat in real time. Speech is picked up on your server, transcribed, matched against rule packs, and acted on per severity — a chat warning or a timed voice mute — while staff receive the player, severity and transcript.

It is a Paper plugin. Chat filters cannot see voice chat; this closes that gap without asking a human to listen to every conversation.

How a flag is made

  1. A player speaks and Simple Voice Chat hands the plugin their microphone packets.
  2. Voice activity detection cuts the stream into utterances using the capture settings.
  3. The selected engine transcribes each utterance.
  4. The transcript is matched against the rule packs for the languages you run.
  5. A match becomes the action configured for that severity, plus a staff alert.

Audio is analysed in memory and never written to disk by the plugin. An utterance goes only to the processor you configured. See consent and privacy.

What it detects

Version one focuses on slurs and fixed-phrase harassment, including English and Czech rule packs. Rules are phrases and patterns, not bare keywords, so ordinary game phrases do not trip a harassment rule.

Full contextual threat reading is not in this version. Read severities and actions before choosing what staff should do with a match.

Copied