TommygunNPCVoices

NPCs that talk back — in any voice you want. Clone any voice from a WAV sample, record your own lines, or use built-in TTS voices. Drop audio files in a folder and your NPCs start speaking with them immediately. Players talk to NPCs through proximity voice chat and get real spoken responses — full speech-to-text, LLM conversation, and text-to-speech pipeline injected through Rust's native voice system. No mods, no client-side anything.

Want your NPCs to sound like a grizzled drill sergeant? Record a sample clip, drop it in, done. Want them to sound like your server's most notorious player? Grab a recording of their voice and now every scientist on the map talks like them. The plugin auto-discovers WAV and MP3 files, pre-encodes them to Opus, and serves them with zero latency. Other players nearby hear the NPC talking in proximity voice chat, just like a real player.

Why This Is Different

• Real voice conversation — not chat messages or pop-up text. NPCs speak through Rust's proximity voice system. Other players nearby hear it too.
• Full pipeline — STT captures player speech, LLM generates a response, TTS speaks it back. All automatic.
• Custom voice library — drop your own WAV files in a folder and NPCs use them. Record your own voice lines, rip audio from movies, clone voices externally — anything goes.
• Per-NPC personalities — each NPC archetype has its own personality, knowledge, voice, and behavior triggers
• $2-3/month — recommended stack (Groq STT + Groq LLM + Voxtral TTS) costs next to nothing for a typical server.

Audio Library

Drop WAV or MP3 files into data/TommygunNPCVoices/audio/ and NPCs start using them. The plugin auto-discovers files, pre-encodes to Opus, and caches everything in memory for instant playback. Use this alongside the live TTS pipeline — pre-recorded files for barks and combat callouts, TTS for dynamic conversation.

How to build your NPC voice library

• Record yourself — grab a mic, record aggro barks, death screams, taunts. Drop the WAVs in. Your NPCs now sound like you.
• Use any voice tool — ElevenLabs, Bark, Piper, whatever. Generate lines externally, save as WAV, drop them in.
• Rip from anything — movie clips, game audio, sound packs. If it's a WAV or MP3, it works.
• Per-player voice capture — record a player's voice on your server, use those clips as NPC voice lines. Your regulars become the NPCs.

Organize by event

Structure your audio folder by trigger type and files auto-map to events:

audio/
  combat/aggro_1.wav
  combat/aggro_2.wav
  combat/aggro_3.wav
  death/last_words_1.wav
  death/last_words_2.wav
  taunt/kill_taunt_1.wav
  ambient/creepy_hum.wav
  trader/greeting_1.wav

Multiple files per trigger = random selection. Five aggro barks means each fight sounds different.

LLM-powered audio selection

The LLM knows your entire audio library. When responding to a player, it can embed [PLAY:filename] tags to mix pre-recorded audio with live TTS:

[PLAY:evil_laugh.wav] You think you can raid MY base?

The laugh plays instantly from cache (zero latency), then the text is spoken via TTS. Lower cost, more expressive, faster than pure TTS.

Vocal sound tags

The LLM can inject natural sounds mid-sentence: [cough], [laugh], [sigh] — making NPC speech feel alive, not robotic.

Admin audio tools

The Voice Pipeline

• Player speaks — proximity voice is captured and buffered with silence detection
• Speech-to-text — audio sent to Groq Whisper (or Deepgram, Voxtral) for transcription
• LLM thinks — transcript + conversation history + NPC personality + situational context sent to Groq LLM
• Text-to-speech — response synthesized by Voxtral (or Groq Orpheus, OpenAI, Deepgram)
• Voice injection — audio encoded to Opus and broadcast through Rust's native voice system from the NPC's position

NPC Archetypes

• Per-NPC personality — custom system prompt with personality traits and domain knowledge
• Per-NPC voice — assign different TTS voices to different NPC types
• Conversation memory — NPCs remember what you said (configurable history length). Admin-injected memories persist across restarts.
• Initiation — NPCs can speak first when a player walks into range. Configurable phrases or LLM-generated greetings.
• Fourth wall toggle — NPCs can be aware they're in a game, or not
• Situational awareness — LLM context includes NPC health, combat state, time of day, weather, player gear, nearby threats, squad count, corpses nearby

Activation modes

Control how NPCs respond. Enable multiple modes per archetype:

13 Voice Triggers

NPCs react to in-game events with voice lines — pre-recorded audio, TTS phrases, or live LLM responses. Each trigger has its own chance roll, cooldown, and audio pool:

TTS Providers

Rate Limiting

• Per-NPC cooldown — prevents spam from a single NPC
• Nearby radius cooldown — stops 10 NPCs all talking at once in crowded areas
• Per-trigger cooldown — each event type has its own timer
• Chance rolls — triggers fire probabilistically, not every time
• Per-player rate limit — configurable API request cap (default 10 req/60 sec per player)
• Busy gate — NPC waits for current response to finish before accepting new input

Admin Controls

• In-game settings panel — full CUI with tabs for general, providers, archetypes, and status
• Live archetype editing — create and configure NPC personalities without restarting
• Conversation monitor — real-time feed of all NPC conversations with timing breakdown
• NPC browser — list all tracked NPCs, view conversation history, teleport-to-NPC
• Debug tools — voice packet capture, replay, WAV export, sine wave test, STT test mode

Commands

Compatibility

Works with any NPC that extends BasePlayer — vanilla or modded. Archetypes are matched by NPC display name, so if the name matches your config, the NPC gets a voice.

• All vanilla scientists — military tunnel, oil rig, junkpile roamers, excavator, arctic base, etc.
• HumanNPC — custom NPCs created with HumanNPC get full voice support
• BotReSpawn / BetterNPC — spawned scientists and custom bots
• Raidable Bases — NPC defenders
• Any plugin that spawns named NPCs — if the NPC has a display name that matches an archetype, it works

Requirements

• Concentus.dll — managed Opus codec, download here and place in RustDedicated_Data/Managed/
• API keys — recommended: Groq (STT + LLM) + Mistral (Voxtral TTS). ~$2-3/month for a typical server.

Permissions

Plugin API (for developers)

Other plugins can use the full voice pipeline without building it:

Hook:

Support

• For all questions and feedback, dm me (1928tommygun) on my discord at discord.fragmod.com

License

TOMMYGUN'S EULA – BY USING THIS PLUGIN YOU AGREE TO THE FOLLOWING!

• Code contained in this file is not licensed to be copied, shared, resold, or modified in any way.
• You may copy the plugin freely to each server instance that your organization owns.
• Do not share this plugin with other server organizations — they must purchase their own licenses.