Krafton said on the 29th that it released the voice artificial intelligence (AI) foundation model "A.X K2 Raon-Speech" on the AI development platform Hugging Face.
This model is one of the second outcomes of the independent AI foundation model project led by the Ministry of Science and ICT.
A.X K2 Raon-Speech is a voice language model with 21 billion (21B) parameters. According to Krafton, among open voice language models with 30 billion parameters or fewer, it ranked No. 1 in overall Korean performance and No. 3 in overall English performance.
In particular, based on SK Telecom's small language model "A.X K2," Krafton combined a self-trained voice encoder and voice codec. Krafton said it developed and applied a training method that stably links voice input and output while maintaining the language model's text capabilities.
In Apr., Krafton launched the AI model brand "Raon" and open-sourced four models: the voice language model "Raon-Speech," the real-time voice conversation model "Raon-SpeechChat," the text-to-speech model "Raon-OpenTTS," and the vision encoder "Raon-VisionEncoder."
Lee Kang-uk, Krafton's chief AI officer (CAIO), said, "A.X K2 Raon-Speech is the result of Krafton's continued research and advancement of AI foundation technology," adding, "We will continue to advance AI foundation technology needed for games and create new gaming experiences."