Emo is an expressive AI chatbot built with Nuxt on the frontend and Lemonade Server on the backend. It transcribes your speech in real time and dynamically changes its facial expression based on the AI's response.
Features:
- Runs entirely on your local machine via Lemonade Server — no data sent to the cloud
- Real-time speech recognition and transcription
- Dynamic emotion display driven by AI responses
- Thanks to Lemonade Server, you can use a GPU (NVIDIA/AMD) or a Ryzen AI NPU (AI 300 series and later) for better performance
Limitations:
Due to the models currently supported by Lemonade Server, the following constraints apply:
- English only — The TTS model (
kokoro-v1) is trained on English. - Speech recognition latency — Whisper currently runs on CPU only, so it can be hard to achieve both low latency and high accuracy at the same time
npm install
cp .env.example .envLaunch Lemonade Server and load the following models:
| Role | Model |
|---|---|
| LLM | Gemma-3-4b-it-GGUF (or any model suited to your environment) |
| Speech Recognition | Whisper-Base |
| TTS | kokoro-v1 |
If you use different models, update the values in .env accordingly.
npm run devOpen http://localhost:3000/ in your browser.
WSL networking issue: If you run Lemonade Server on Windows and Emo inside WSL, Emo may not be able to reach the server. To fix this, open WSL Settings and set Networking Mode to Mirrored.
- Make Emo recognize its own name and the pronunciation.
