Jack Dorsey’s Block showed Dessa, a new voice model from its research lab that talks with laughs, pauses, and tears, TechCrunch and The Verge reported. Block says Dessa is trained from scratch on human speech only, not on text pipelines that flatten emotion.

Dessa is full-duplex, meaning it can talk and listen at the same time, like OpenAI’s GPT-Live model, TechCrunch said. Knowledge comes from the conversational audio it was trained on, not a separate large language model bolted on for facts. The company says it ran training on a single NVIDIA GPU rack. Emotion is built into the model rather than steered with voice tags, Block claims. "Talking with it feels like talking with a person," the company said in a post shared by TechCrunch.

Demo clips show awkward small talk, a guessing game, a Christopher Nolan chat that wanders into Superman films, and a moment the site describes as teary gratitude on a surprise Paris trip, The Verge reported. Block says interacting with today’s voice AI "can just plain suck" because it turns speech into text and loses feeling that matters for safety when a user is in pain.

The model can run locally on a MacBook Pro with an M4 Max chip or better. There is no public download yet. Block is inviting a small group of AI safety researchers for early access, TechCrunch and The Verge reported. Square already uses voice AI for ordering; an on-device model could cut cost for more business tools.

TechCrunch said the demo’s emotional range stood out most when Dessa laughed mid-sentence or paused before answering a hard question. Block frames the project as research, not a consumer product launch.

What we know: Dessa is a local, full-duplex speech model in research preview. Reviewers find it strikingly human.
What we don't: When, or whether, Block will open wider access.