Eleven Labs unveiled its voice-to-speech (S2S) model "Eleven Real-time" on the 3rd, according to the global AI audio company. The model combines speech recognition, turn conversion, and speech synthesis into a single system, and supports more than 90 languages including Korean and Japanese. It answers in 0.57 seconds after a question is over. The models are divided into two types: "Eleven Real Time v1 Mini," focused on fast and natural conversations for AI assistants, educational tutors, and game and entertainment characters, and "Eleven Realtime v1 Max," focused on complex multi-stage tasks such as customer support, selecting potential customers, and calling external tools. Both models support tool calling to external programs or services. According to last month's evaluation by AI model evaluation agency Artificial Analytics, v1 Max recorded a 98.8% correct answer rate in "Big Bench Audio," which measures speech-based reasoning ability. Both models support more than 17,000 voice libraries owned by Eleven Labs.