New modelhigurashi-ai
Haru-4B
An open speech model that listens and speaks at the same time, as on a phone call, in nine languages. A text card sets its character and a short clip sets its voice.
What’s new: For researchers building voice agents. It is a base model under Apache 2.0 that runs on an Nvidia GPU with about 12 GB. It does not follow instructions like an assistant and is not a reliable source of facts.
298 likes on Hugging Face