Core Dialog – Local NPC Talk AI

🗣️ Core Dialog: Advanced NPC Talk AI

Bring your Unity projects to life with Core Dialog, a fully local, zero-API AI communication framework. Designed for seamless integration into Unity, this system empowers your NPCs to listen, reason, and speak in real-time, completely offline. 🧠⚡

By leveraging state-of-the-art local inference libraries—Llama.cpp (via LLamaSharp), Whisper.net, and ONNX Runtime—Core Dialog bypasses expensive cloud AI costs and latency. From context-aware shopkeepers to emotionally reactive companions, this complete pipeline handles Speech-to-Text (STT), Large Language Model (LLM) processing, and Text-to-Speech (TTS) natively inside your Unity environment.

⚙️ Technical Specifications & Dependencies

Core Dialog is built on top of robust, open-source AI frameworks. The asset includes the necessary pre-compiled DLLs and wrappers to make these run natively within Unity.

• 🦙 Large Language Model (LLM): Powered by LLamaSharp (a C# wrapper for llama.cpp). Supports GGUF format models (Llama 2, Llama 3, Mistral, etc.). Highly optimized for both CPU and GPU inference depending on the hardware.
• 🎙️ Speech-to-Text (STT): Integrates Whisper.net (based on whisper.cpp) for fast, local voice recognition. Capable of real-time transcription directly from the Unity Microphone class.
• 🔊 Text-to-Speech (TTS): Uses Pocket TTS driven by the Microsoft ONNX Runtime library for rapid, low-latency audio synthesis, including rapid 5-second voice cloning.

✨ Core Features

• 🚫🌐 Offline Real-Time Processing: No API keys, no internet connection required, and zero subscription costs. Everything runs on the user's local hardware.
• ⌨️🎤 Dual Input Modes: Supports standard UI text input as well as Voice Activity Detection (VAD). Speak into the microphone, and the system automatically detects when you stop speaking to process the STT.
• 🌍 Context & World Awareness: Feed the LLM real-time data about the player's inventory, health, or location. NPCs can perform actions—like a shopkeeper searching their inventory to recommend a weapon based on the player's current stats.
• ❤️😡 Emotional & Relationship Systems: Built-in narrative variables allow NPCs to track "Trust" or "Anger." An enemy will grow hostile if insulted, while a companion will unlock new dialogue branches as trust increases.
• 👥 Voice Cloning: Generate custom NPC voices instantly by providing a single 5-second .wav file of your desired voice.

💻 Hardware Requirements & Performance

Because this asset runs complete AI models locally, performance scales with the player's hardware.

• Minimum Specs (CPU Inference): 16GB System RAM, modern multi-core CPU (Core i5/Ryzen 5 or better).
• Recommended Specs (GPU Inference): 4GB+ VRAM (NVIDIA GTX1650Ti or better), 16GB System RAM
• Unity Compatibility: Unity 6000.3 LTS or newer.

🖥️ Supported Platforms

• 🪟 Windows (x64): CPU | GPU - Hardware-accelerated
• 🍎 macOS:support for both Intel and Apple Silicon (M1/M2/M3) chips.
• 🤖 Android: Native ARM64 support.
• 📱 iOS: Coming Soon!

📦 Included Content & Demo Scenes

The package includes highly documented code, modular prefabs, and several demo scenes to demonstrate specific use cases:

• 🏪 The Shopkeeper: Demonstrates an NPC querying a simulated inventory system to make contextual recommendations.
• 💕 The Romantic Companion: Showcases emotional trust variables and conversational memory tracking.
• ⚔️ The Hostile Enemy: Highlights narrative triggers where the NPC gets angry based on player insults, shifting their behavior state.
• 🤖 General Chatbot & Voice: A clean, UI-only scene showing the raw STT to LLM to TTS pipeline.
• 🧪 Voice Cloning Sandbox: A utility scene to test and generate NPC voices using short audio clips.