Inworld Launches TTS-2, Outperforming Sonic3.6 in Voice Cloning

ME News report, September 3 (UTC+8), Beating AI Quick News: Real-time voice AI company Inworld has officially launched Realtime TTS-2. In the latest Controlled Voice Arena by Artificial Analysis, it ranks first with an Elo score of 1123, surpassing Cartesia Sonic 3.6’s 1119 by 4 points. This leaderboard pits different models against each other by having them clone the same set of eight voices and then conducting blind tests to determine which sounds more natural and human-like, evaluating factors such as pronunciation, pauses, intonation, and rhythm. Realtime TTS-2 incorporates previous segments of real speech as context, analyzing the user’s tone, rhythm, and emotional state to adjust its next response accordingly. Developers can also directly input natural language instructions such as “speak softly” or “sound tired but warm after a long day,” and incorporate sounds like laughter, sighs, and breathing. It also supports maintaining the same voice across languages and cloning voices using short audio samples. Inworld states that this generation of the model is specifically designed for real-time conversation, leveraging intonation, rhythm, and emotional states within full voice interactions. (Source: BlockBeats)