Deepie - Voice Recognition & Local AI Horror Game (Preview Build)

Deepie is a next-generation atmospheric horror game that pushes the boundaries of player-NPC interaction by integrating advanced Local Large Language Models directly into the game loop. Built on Unity HDRP, the game delivers an immersive psychological horror experience where the environment isn't the only thing reacting to you—the entity is listening.

💾 🚨 CRITICAL INSTALLATION & EXTRACTION GUIDE (READ BEFORE PLAYING)

To bypass platform file size limits and ensure the game runs without system errors, please follow these exact steps:

1. Download Both Parts: You must download both Part 1 (Main Game) and Part 2 (StreamingAssets - AI Model) from the download section below.

2. Extract Part 1 First: Extract Part 1 (Main Game) to a local drive folder (e.g., C:\Games\Deepie or D:\Deepie).

3. Install the AI Model (Part 2): Place the streamingassets folder directly into the Deepie/Deepie_Data directory.

4. ⚠️ DO NOT EXTRACT UNDER ONEDRIVE: Windows OneDrive folders can block the local AI files from loading. Make sure your installation path is on a standard local drive.

5. Windows Defender Notice: Since the local AI model communicates directly with the Unity loop, Windows Defender might flag the game. If the game crashes on startup or the AI doesn't talk, add the game folder to your antivirus exclusion list.

⚠️ HARDWARE & SYSTEM REQUIREMENTS

Because the game runs a massive 4 Billion Parameter Language Model completely locally on your hardware, it has strict system requirements. No Internet/Cloud APIs are used.

• Graphics Pipeline: Unity HDRP (High Definition Render Pipeline)

• Target Hardware: Mid-to-High Tier Gaming PCs

• Minimum VRAM: 3-4 GB Dedicated VRAM (6 GB+ Recommended for smooth performance)

• Storage: ~3.50 GB (Includes encapsulated local model layers)

🧠 CORE TECHNOLOGICAL SYSTEMS

1. Zero-Cloud Local LLM Integration (Qwen 3.5 4B)
Unlike traditional AI games that rely on high-latency cloud APIs (like OpenAI), Deepie executes a 4B parameter Large Language Model fully offline using Unity Sentis. The entity processes data locally on your GPU, eliminating connection lag and ensuring uninterrupted, adaptive behavioral patterns.

2. Real-Time Voice Interaction Loop (STT / LLM / TTS)
The core gameplay revolves around a dynamic acoustic loop. The game captures your microphone input in real-time via local Speech-to-Text (STT), feeds it directly into the entity’s cognitive prompt layer, and streams the output back through a local Text-to-Speech (TTS) synthesizer connected to Unity’s 3D Audio Spatializer.

3. High-Fidelity HDRP Atmospheric Horror
Leveraging the power of Unity’s High Definition Render Pipeline, Deepie features advanced volumetric lighting, physically-based materials, and dark, claustrophobic environments optimized to sustain heavy VRAM sharing between the graphics pipeline and the neural model layers.

🛠️ DEVELOPMENT & COMMUNITY FEEDBACK

This project is an engineering showcase exploring the fusion of local edge-AI and modern game design. Since managing VRAM allocation between HDRP and a 4B local model stretches consumer hardware to its absolute limits, player feedback regarding performance, latency, and hardware specs is highly valuable.

Please report your GPU specs and voice-response latency in the comments below to help optimize future builds.

#horror #action #adventure #llm #scary #psychological_fear #ai #tts #stt #monster #microphone