Krafton launches the artificial intelligence (AI) model brand Raon. /Courtesy of Krafton

Krafton said on the 2nd that it unveiled its artificial intelligence (AI) model brand "Raon."

Raon is a name inspired by the pure Korean word meaning "joy," and the company said it reflects Krafton's philosophy of creating the essential joy of games through AI technology.

The models unveiled this time are the speech-enabled large language model (LLM) "Raon-Speech," the real-time voice conversation model "Raon-SpeechChat," the text-to-speech (TTS) model "Raon-OpenTTS," and the vision encoder (image feature extraction model) "Raon-VisionEncoder."

Krafton released the four models covering voice and visual information as open source on Hugging Face, the world's largest AI model platform. A company official said, "We will strengthen our global AI technology competitiveness with the Raon AI model brand at the center going forward."

"Raon-Speech" is a speech language model capable of speech understanding and generation, with 9 billion (9B) parameters. According to Krafton, a comprehensive evaluation of seven core tasks—such as speech-to-text conversion, text-to-speech conversion, and speech-based question-and-answer—and 40 benchmarks found that among open speech language models with 10B parameters or fewer, its English and Korean performance ranked No. 1 globally.

"Raon-SpeechChat" applies real-time bidirectional communication technology that allows users and the model to freely interrupt during conversation. The company said it is the first real-time bidirectional voice model announced in Korea.

"Raon-OpenTTS" is a text-to-speech model trained only on open speech data. "Raon-VisionEncoder" converts images into information that AI can understand, and its hallmark is that it was trained from scratch in-house without using a pretrained model, relying only on open data.

Lee Kang-uk, Krafton's chief artificial intelligence officer (CAIO), said, "By sharing large-scale training data and core models as open source, we hope researchers and developers can use them freely and that this contributes to the advancement of multimodal technology and the growth of Korea's AI ecosystem."

※ This article has been translated by AI. Share your feedback here.