A.X K2 RAON-SPEECH · 21B-A3B
Your voice,
understood.
One bilingual speech model for transcription, audio understanding, and natural speech generation in English and 한국어.
21.2Btotal parameters
3.5Bactive parameters
24 kHzneural audio codec
01 / STUDIO
Try the model
Choose a task. Your request runs on KRAFTON's official public Space.
Transcript
Your transcript will appear here.
Answer
The model's answer will appear here.
40 characters
Voice & continuation settings ＋
Generated speech24 kHz
Generated audio will be ready to play and download here.
Speech in.
Intelligence out.
Raon-Speech unifies an AuT speech encoder, a Mimi-style neural codec, and a mixture-of-experts text backbone in one multimodal model.
- 01Bilingual English & Korean
- 02Speech recognition & understanding
- 03Natural TTS & speaker conditioning
- 04Paralinguistic awareness