GitHub radar
Breeze TTS 2: open voice cloning, #1 on benchmark
BreezeBlue released Breeze TTS 2 — an open-weight TTS model with voice cloning, voice design from text descriptions, and emotion control, ranking #1 among open-weight models on the Artificial Analysis TTS leaderboard.
Breeze TTS 2 is an open-weight text-to-speech model from BreezeBlue, ranking #1 among open-weight models on the Artificial Analysis TTS leaderboard and claiming to outperform several closed proprietary systems. It has 3.5 billion parameters and supports three use modes: Voice Clone (reproduce any voice from a short audio reference), Voice Design (describe a voice in natural language — age, accent, style — and the model generates one), and Voice Direction (clone a voice with emotional steering). The model also supports Vocal Events: expressive sounds like laughs, coughs, and sighs can be embedded directly in the synthesis transcript. Latency to first audio is under 40ms, making it suitable for real-time conversational use. Supported languages are English and Chinese. Requires a CUDA-capable NVIDIA GPU (12 GB for standard inference, 24 GB for fast mode), Linux, and Python 3.10+.
Why a vibe-coder should care
For English-language content creation — videos, narrations, podcasts, or interactive voice demos — Breeze TTS 2 delivers voice cloning and real-time synthesis at a quality level that previously required paid commercial APIs. Local deployment means no per-character cost and no data leaving your machine. Note: Russian language is not supported — for Russian speech synthesis a different model is needed.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Install Breeze TTS 2 from BreezeBlue: https://huggingface.co/BreezeBlue/Breeze-TTS-2 — follow the README, load the model, and synthesize 'Hello, this is a test of Breeze TTS 2' using voice cloning from a short audio reference. Ask me if you need a reference audio file.
Requires an NVIDIA GPU with 12 GB of VRAM or more and Linux — a MacBook won't work.
Open on Hugging Face▌ More finds