handcrafted-persona-engine

Why we included this project

Streamers who want a character that actually talks back instead of just answering in a chat box get a complete local pipeline here, wrapped in one desktop app. It runs microphone audio through Whisper for speech recognition, feeds the transcript to an LLM steered by a personality file, and returns the reply as real-time TTS, with optional RVC voice cloning and Live2D lip sync that can render in a transparent overlay or pipe into OBS over Spout. The bundled Aria model is rigged and ready, and you can bring your own Live2D model. Expect Windows plus an NVIDIA GPU, and a multi-gigabyte model download on first launch, though the in-app installer handles profile selection, hash checks, and repairs. If you want an AI co-host, a stream mascot that reacts to chat, or an animated desktop assistant that speaks, this is a solid reference and a fast starting point.