KRAFTON releases 1B zero-shot voice-cloning TTS

KRAFTON has published Raon-OpenTTS-1B, a one-billion-parameter text-to-speech model that generates English speech and can mimic a target voice from a short reference sample. The release is available on Hugging Face and marks the first version of the Raon open TTS line.

The model is a flow-matching diffusion transformer, an architecture that has become popular for high-quality audio and image synthesis because it can produce natural output in relatively few sampling steps. Positioned as a zero-shot voice-cloning system, it aims to reproduce a speaker's timbre without per-voice fine-tuning, relying instead on a reference clip at inference time.

What it offers

• Roughly 1B parameters, a size that stays practical to run on a single modern GPU
• Zero-shot voice cloning from a short reference sample
• English-language output
• A flow-matching diffusion-transformer backbone

Why it matters

Open zero-shot TTS remains a competitive space, and a compact 1B model that clones voices lowers the barrier for developers building narration, accessibility, and assistant features without training custom voices. The tradeoff is scope: this initial release targets English only, and the repository carries a non-standard "other" license, so teams should check the terms before commercial use. As a version 1.0, it establishes a baseline the family can iterate on.