Krafton's A.X K2 Raon-Speech AI: New Voice Tech for Games

Krafton has released its new voice artificial intelligence model, 'A.X K2 Raon-Speech', on the global AI platform Hugging Face. This model is a key outcome from a project led by the Ministry of Science and ICT. Here's the thing: It combines SK Telecom’s small language model with Krafton’s own voice encoder and codec. What's interesting is this system can directly convert speech between voices. It tackles a limitation where emotion and intonation were lost in previous AI speech-to-text conversions. Krafton trained its own voice encoder and codec to reproduce a person’s tone and emotion. The model has 21 billion parameters. It ranked first in overall Korean performance and third in English among publicly available speech language models of 30 billion parameters or less. It underwent 46 benchmark evaluations in areas like speech recognition and speech synthesis. Krafton plans to use this voice AI to advance its in-game companion characters. This means AI characters could understand a user’s voice and intent in real time, interacting with situation-appropriate speech. This could create new game experiences for players.