How does Block’s new Dessa voice model sound so human, and can you use it yet?

Block’s Dessa, unveiled October 23, is a full-duplex voice model that listens and speaks at the same time, like a real phone call — and its lab says it learned conversation from recorded human speech alone, with no text and no pretrained model underneath. There is no public access yet, and Block has released recordings rather than benchmarks, so the “eerily human” claims are still untested.

Short answer: Dessa is a voice model from Block — the company behind Square and Cash App — unveiled on October 23. It sounds human, its research lab says, because it learned conversation the way people do: by listening to recorded human speech. Unusually, it was trained on speech alone, with no text and no pretrained model underneath.

And no, you cannot use it yet — there is no public access, only recordings Block has chosen to release. The key phrase is “full-duplex.” Most voice assistants take turns: you talk, there is a pause, it talks, like a walkie-talkie. Full-duplex means both directions run at once, like a real phone call — the model can listen while it speaks, react mid-sentence, and handle the little overlaps and “mm-hm” sounds that make conversation feel alive instead of scripted.1.

How it learned: from raw recordings of people talking, picking up rhythm, timing, and turn-taking directly from audio rather than from written transcripts.2. What it runs on: a MacBook Pro with an M4 Max chip or better — serious hardware, but a laptop, not a data centre, which hints at private, on-device voice one day.3. What we do not know: Block has published recordings, not independent benchmarks.

Hand-picked demos show a model at its best; they do not tell you how it handles accents, noise, or a bad phone line.Pro tip: Never judge a voice model by its launch recording. Judge it by the first time you can interrupt it, mumble at it, and talk over a barking dog — that is the test real conversation sets.Ad’s take: TechCrunch called Dessa “eerily human,” and the recordings are impressive. But “eerily human in a demo” and “trustworthy on your phone” are two different products.

I will believe the second one when I can try it myself.