Blue TTS v3: open-source TTS, about 50x real-time on a CPU
retained text
Blue TTS v3 is out, an open-source TTS model that runs locally. Almost everything changed since v2. Speech is more natural and expressive, with a **MOS of 4.43 out of 5**. Voice cloning has improved significantly, and the model is fast and lightweight: it runs locally at **50x real-time on a CPU**, no GPU needed. MIT license, 388 MB, 44.1 kHz output, 7 languages: English, German, Spanish, Italian, Russian, Hebrew, Yiddish. * Code: [https://github.com/maxmelichov/BlueTTS](https://github.com/maxmelichov/BlueTTS) * ONNX weights (v3): [https://huggingface.co/notmax123/BlueV3-onnx](https://huggingface.co/notmax123/BlueV3-onnx) * PyTorch weights (v3): [https://huggingface.co/notmax123/BlueV3](https://huggingface.co/notmax123/BlueV3) * Live demo: [https://huggingface.co/spaces/notmax123/BlueV3](https://huggingface.co/spaces/notmax123/BlueV3) * Website: [https://bluetts.com](https://bluetts.com) Feedback is welcome.
document fetched 2026-11-18 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1xytrz9/blue_tts_v3_opensource_tts_about_50x_realtime_on