3.3k likesNew · 7d old
Haru-4B lets AI talk and listen simultaneously, like a phone call
What is Haru-4B?
The model processes a slice of your speech every 133 ms and instantly generates its own speech and text, enabling continuous two-way conversation. You can choose a character and upload a short voice clip to control how it sounds, in many languages.
Who it is for
Developers building voice assistants, call-center bots, or language-learning tools can experiment with real-time AI dialogue.
How to try it
python -m haru.server --hf-repo higurashi-ai/Haru-4B
The story so far
Early Signal first spotted higurashi-ai/Haru-4B on November 19, 2026, with 3.3k likes. On its busiest day it gained 477 likes. The project is 7 days old. Early Signal files it under Voice, LLMs, Coding.
Questions about Haru-4B
- How do I install Haru-4B?
- Use one of these, depending on your setup: git clone https://github.com/higurashi-ai/Haru && cd Haru · pip install -e ./haru · python -m haru.server --hf-repo higurashi-ai/Haru-4B
- What do I need to run Haru-4B?
- NVIDIA GPU with ~12 GB VRAM and Python