# Breeze TTS 2

> Self-hosted open-weight TTS model with voice cloning, design from descriptions, and voice direction. Streaming REST API and CLI for real-time synthesis on NVIDIA GPUs.

- **Category:** AI
- **Licence:** Apache-2.0
- **Open source:** open_core
- **Languages:** Python
- **Platforms:** Docker, pip
- **Tags:** API, Real Time, Pytorch, Streaming, Cuda, Text To Speech, Huggingface, Voice cloning, Nvidia Gpu, Voice Design, Voice Direction

## Description

An open-weight text-to-speech system for real-time interaction that leads the Artificial Analysis leaderboard among open models. Self-host via Docker, pip, or Python. Three synthesis modes: clone a speaker from reference audio, design a voice from a text description, or direct a cloned voice with instructions. Inline vocal events add laughter, sighs, and more. A streaming HTTP API delivers 24 kHz PCM on port 7860; a batch endpoint processes up to 128 texts. Requires an NVIDIA GPU with 12 GB VRAM (eager) or 24 GB (fast path). Weights on Hugging Face under a non-commercial license; code is Apache-2.0. Commercial use needs a paid BreezeBlue subscription.

## Links

- [Homepage](https://github.com/breezeblue-ai/breeze-tts)
- [Website](https://breezeblue.ai/)
- [GitHub Repository](https://github.com/breezeblue-ai/breeze-tts)
- [Hugging Face Model](https://huggingface.co/BreezeBlue/breeze-tts-2)
- [Blog Post](https://breezeblue.ai/breeze-tts-2)
- [Discord Community](https://discord.com/invite/6H7AgPe9pA)
- [X (Twitter)](https://x.com/BreezeBlueX)

---
Source: https://selfhost.directory/project/breeze-tts