KRAFTON releases Raon-OpenTTS-1B, an open zero-shot voice cloning model
victormustar · x · 2026-08-20
KRAFTON has released Raon-OpenTTS-1B, an open-weight zero-shot Text-to-Speech model for voice cloning. It achieves the best overall similarity score and lowest error rate compared to leading TTS models, matching the performance of state-of-the-art closed-data systems.
Key Features:
- Fully Open: Both model weights and training data (615K hours) are publicly available.
- Competitive Performance: Ranks 1st or 2nd in WER and SIM on major benchmarks like Seed-TTS-Eval.
- Robust: Delivers top performance across clean, noisy, wild, and expressive acoustic conditions.
- Large-Scale Data: Trained on a curated dataset of 510K hours.
More from Multimodal
- Dreamina launches Seedream 5.0 Lite AI image model — aftahi_ai · 2026-08-20
- Testing LTX 2.5: Using Reference Images for Video Gen — cocktailpeanut · 2026-08-20
- AI Short Film 'PRIME CUT': A Crime Fiction Creation from 4 Months Ago — TheSamsaBoy · 2026-08-20
- New LTX-Video LoRA generates scenes from new angles — makeitrad1 · 2026-08-20
- Seeking Advice: Flux Dev Face Consistency Issues and Alternatives for Realism — bigM1232 · 2026-08-20
- MiniMax H3 Issues: Cropping, Camera Movement, Ignoring References — Silver-Spot-2763 · 2026-08-20