Breeze-TTS-2: A New Frontier-Quality Local Text-to-Speech Model
Breeze-TTS-2, a new open-weights text-to-speech (TTS) model developed by BreezeBlue, has been released with a model size of approximately 7GB. It is available for local deployment and features an interactive demo playground on Hugging Face. The model brings "frontier-level" voice generation quality to local hardware, reducing reliance on proprietary cloud APIs. It ranks highly among open-weights TTS models, offering advanced capabilities like bilingual support, voice cloning, and voice direction. Breeze-TTS-2 ranks third among open-weights TTS models with an Elo score comparable to Fish Audio S2 Pro. The model supports bilingual text-to-speech, voice cloning, and voice direction, and can be run locally with a ~7GB footprint.
## BACKGROUND
Text-to-speech (TTS) technology has shifted from robotic synthesis to highly realistic, expressive voice cloning using deep learning. While top-tier models are often locked behind commercial APIs, the open-source community actively develops "open-weights" models that users can run privately on their own hardware.