MNN-TaoAvatar Android - Local 3D Avatar Intelligence
Welcome to MNN-TaoAvatar App, a full-featured mobile app about the TaoAvatar paper (arXiv:2503.17032v1). This project brings multimodal AI avatars to life directly on Android devices, running all models locally, including:
- LLM (Large Language Model)
- ASR (Automatic Speech Recognition)
- TTS (Text-to-Speech)
- A2BS (Audio-to-BlendShape)
- NNR (Neural Rendering)
The iOS App will be coming later, stay tuned for updates!
Features
- Conversational AI powered by a local LLM
- Speech-to-text with embedded ASR models
- Voice synthesis with TTS on-device
- Avatar behavior animation via A2BS(Audio-to-BlendShape)
- Real-time neural rendering for expressive avatars
- 100% offline and privacy-focused
How it Works

Requirements
Because all AI models are executed locally on-device, this project requires high-performance hardware to run smoothly.
Minimum Device Requirements
- Snapdragon 8 Gen 3 or equivalent flagship SoC
Examples: Snapdragon 8 Gen 3, Dimensity 9200 to have smooth experience.
- 8 GB RAM or more
- 5 GB free disk space for model files
- ARM64 architecture
⚠️ Devices below these specs may experience lag, audio stutter, or limited functionality.
Setup and Run
1. Clone the Repository
git clone https://github.com/alibaba/MNN.git
cd apps/Android/MnnTaoAvatar
2. Build and Deploy
- Connect an Android device
- Hit Run in Android Studio or use:
./gradlew installDebug
Releases
Version 0.0.1
- Click here to download
- this is our first public released version; you can chat with 3d avatar in the app with asr and tts; if you have any questions, please feel free to open an issue for assistance.
Links And References