Block Announces Dessa, A Voice AI Trained Only On Conversational Speech
Dessa is a voice model built from speech rather than text, and the gap between the announcement and a working release is now occupied by an AI safety review, not a launch date.
Reporting from 1 source: GIGAZINE.
Block, the company led by CEO Jack Dorsey, announced Dessa, a voice conversation model trained only on human speech rather than text. Its development team says Dessa responds while a person is still speaking, chimes in, and laughs at utterances. It was trained on a single NVIDIA GB200 NVL72 and can run locally on a MacBook Pro with an M4 Max. Neither the API nor the model has been released, and Block is recruiting AI safety researchers.
The team behind Dessa came out of a Toronto startup of the same name that has researched AI since 2012, and now works as a small research group inside Block. Its results go directly to Dorsey. The announced model handles conversation differently from voice assistants built on text LLMs: it can overlap a speaker, interject, and laugh in response, rather than waiting for a pause. In one demonstration it helps a user recall a game title by offering "Majora's Mask?" mid-conversation. What exists today is a demonstration and a safety hiring call.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google