KRKrafton, a South Korean video game developer, has launched a new audio foundation model called A.X K2 Raon-Speech, which aims to enhance the realism of in-game characters' voices. This model is part of Krafton's ongoing research within SK Telecom’s consortium working on the Ministry of Science and ICT’s Independent AI Foundation Model project. A.X K2 Raon-Speech integrates Krafton’s proprietary voice encoder and audio codec with a compact language model derived from SK Telecom’s A.X K2. With 21 billion parameters, it ranks highly among both Korean and English-language models with similar parameter counts. The model was evaluated across six domains using 24 Korean and 22 English benchmarks. Unlike many conversational audio systems that convert speech to text and back, A.X K2 Raon-Speech preserves emotional and tonal nuances, aiming to provide more natural-sounding interactions. Krafton intends to integrate this technology into its AI-driven co-playable characters to improve gameplay conversations.
Bias read (Center): The article discusses the development of an advanced AI audio model by a private company for use in gaming. There is no mention of political figures, policies, or contentious issues. The focus is purely on technological innovation and its application in the entertainment industry.
