AI digital people can interact with digital people through voice dialogue, expression control and physical movements. It runs in real time on a laptop equipped with NVIDIA 3060 graphics card.
Project positioning
EdgePersona is an open source project whose goals areCreate a fully locally running, privacy-secure intelligent digital human system。The system has voice dialogue, multimodal action response and other capabilities. It can run on ordinary notebooks and does not require networking to ensure the privacy of user data.
core features
- 100% localized operation: All model reasoning and speech processing are done locally, without the need to connect to the Internet, ensuring privacy and security
- Lightweight and efficient: Low hardware requirements, generally notebooks with NVIDIA GPUs (≥6 GB of memory) can run real-time code
- multi-modal interaction: Support interactive methods such as voice dialogue, character facial expressions and physical action responses.
- Customizable role: Define the character’s personality, tone and action style through YAML profiles.
- Multiple model support: Compatible with local model formats such as Ollama, HuggingFace, and GGUF
Overview of technical architecture
EdgePersona is developed based on Python, and its core modules include:
- Speech recognition (ASR)、Text To Speech Module (TTS)
- Conversation Engine (dialogue.py)、Memory module (memory.py)
- Motion controller (player.py, robot.py) such
- Support calling local LLM (such as the Olama protocol server) for natural language generation
The system integrates multiple open source components, such as Live2D, Ollama, local TTS (such as Kokoro TTS), etc.
how to use
- Clone the warehouse and create a Python virtual environment.
- Installation dependencies:
pip install -r requirements.txt - Configure role and voice parameters (modify YAML, tts.py, model path, etc.)
- Start procedure:
python main.pyYou can run the digital human interactive interface
Example of usage scenarios
- want to build aOffline voice assistant, adapt to personal privacy needs;
- want toOn a laptop or low-profile deviceRun digital human applications;
- Need a system that can customize the character’s voice, movements and dialogue styles;
- pursue aNo need to rely on cloud servicesReady to run AI avatar.
advantages and limitations
| advantages | Limitations/Points of Attention |
|---|---|
| Fully offline, no network dependence, privacy and security | There are requirements for graphics card memory, such as NVIDIA ≥6 GB |
| Lightweight, suitable for ordinary notebooks | The quality of conversation generation is limited by the local LLM model you load |
| Support in-depth customization of character styles | Configuration and model preparation require a certain technical foundation |
| Supports mainstream model formats | Multimodal actions and scene interactions are still under development |
summary
EdgePersona is an open source project dedicated to promoting the “return to localization” of digital people. It abandons cloud dependence and uses local models and local reasoning to achieve a private and customized interactive experience between digital people and users. If you have a device with NVIDIA GPUs and want to build a virtual character or voice assistant that runs offline, deeply customizable, EdgePersona is a very potential starting point.
Github：https://github.com/zc-maker/EdgePersona
Oil tubing: