Block builds Dessa voice AI from human audio alone on a single Nvidia rack

Ragavan Thurairatnam, who heads Jack Dorsey’s Block lab, announced on Oct. 23, 2026, that Block trained a voice model called Dessa using only human speech, with no text or pretrained models. The full-duplex speech-to-speech system talks and listens simultaneously and runs faster than real time, skipping the usual speech-to-text-to-speech pipeline; Block trained it on a single Nvidia GB200 NVL72 rack and says it can run locally on a MacBook Pro M4 Max or better. Dessa is not publicly available. Block is operating a waitlist and giving early access to a small group of AI safety researchers to identify risks the team has not foreseen.