About this project

Hoho Avatar is a source-first toolkit designed to build audio-reactive avatar experiences. It features a browser-based TypeScript engine that analyzes streaming mono PCM audio to drive character animations, specifically managing five mouth states and automatic blinking for layered PNG characters rendered via Canvas 2D. Key capabilities include: - Audio Integration: Supports various audio sources including live microphone input, local audio files, and streaming TTS (via KittenTTS). - Voice Agent Support: Includes a gateway for integration with Azure OpenAI Realtime voice agents for full-duplex conversations. - Asset Management: Uses a specific JSON-based character format for defining body, eye, and mouth layers. It includes a Codex-based 'Skill' for generating and validating these assets. - Modular Architecture: Separates media acquisition, motion analysis (PCMAnalyzer and MouthClassifier), and rendering, allowing for potential future expansion to Live2D or 3D renderers. The toolkit provides a set of example characters and a demo application to showcase real-time voice reaction and local audio playback.