Cartesia ships Sonic-3.6, topping Artificial Analysis text-to-speech rankings

Cartesia/Sonic

Two months after Sonic-3.5, Cartesia rebuilt data, architecture, training and evaluation for Sonic-3.6. It replies in under 90 ms, speaks nearly twice as fast as before, covers 61 locales and ranks first on both Artificial Analysis voice leaderboards.

• Elo 1120 on the controlled-voice board, ahead of Sonic-3.5 (1095) and Eleven v3 (1066).
• Native US-English listeners preferred it over Eleven v3 in 92% of blind tests (company test).
• Adds Odia and Urdu and mixes Hindi and English in one sentence.
• About 132 characters per second, up from 68.