Cartesia ships Sonic-3.6 beta: more natural TTS across 44 languages

Cartesia, a startup specialized in voice synthesis, released Sonic-3.6 in beta, the new version of its model that turns written text into speech. It lands just three months after Sonic-3.5, a sign of how fast release cycles have become in the synthetic voice market.

The headline change is not a new voice but how the model reads. Sonic-3.6 decides on its own where to pause and how to shape intonation based on the context of the sentence, much like a voice actor who reads a script by understanding its meaning. It can also render hesitations like uhm and hmm written in the text, without developers having to insert special markup codes.

For everyday users this means voice assistants, audiobooks and automated call centers that sound less robotic. Supported languages grow from 42 to 44 with the addition of Odia and Urdu, and the model improves its handling of Hinglish (Hindi written in Latin script with English vocabulary) and Indian proper names, a clear signal toward the South Asian market.