Inworld TTS-2 Officially Launched: Simultaneous Interpretation Blind Test Surpasses Sonic3.6 to Take the Lead

Dynamic Beating AI News Flash: Real-time voice AI company Inworld has officially launched Realtime TTS-2. In the latest Artificial Analysis Controlled Voice Arena, it ranks first with a 1123 Elo score, which is 4 points higher than Cartesia Sonic 3.6's 1119. In this ranking, different models clone the same set of 8 voices for a blind test, focusing on who sounds more natural, more human-like, including pronunciation, pauses, intonation, and rhythm.

It will also use previous rounds of real voices as context, listening to the user's tone, rhythm, and emotions to adjust how the next sentence is spoken. Developers can directly write natural language commands like "Speak softly" or "Sound tired but warm as if just off work," and include sounds like laughter, sighs, breaths, etc. It also supports maintaining the same voice across languages and cloning voices from short audio clips.

Inworld officials stated that this generation of models is specifically designed for real-time conversations, utilizing tone, rhythm, and emotional states from complete voice interactions.