Block, the payments company led by Jack Dorsey, has unveiled a voice AI that sounds so human it is splitting reactions down the middle — amazement at how natural it feels, and alarm that it could make phone scams much harder to spot. The model, called Dessa, was trained entirely on human speech rather than text.
Jack Dorsey (Credits: CoinDesk)
Unlike most voice systems, which first convert speech into text before generating a spoken reply, Dessa processes audio directly. That lets it listen while speaking, so it can handle interruptions, laughter and changes in tone mid-conversation. In company demonstrations, Dessa laughs in the middle of a sentence and reacts to a surprise with what sounds like genuine emotion. Block says the conversations were recorded live, without anyone directing the model’s responses.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence,” Dorsey wrote on X.
Block’s Dessa Voice AI: Why It Sounds So Human
Dessa was trained from scratch using a single Nvidia GB200 NVL72 computing system, according to Block. Despite that scale of training, the model is compact enough to run locally on a MacBook Pro equipped with an M4 Max chip or better — no cloud connection required for the demo setup.
That local, speech-only design is part of what makes Dessa stand out. By skipping the text step entirely, it preserves the pauses, emphasis and emotional shading that text-based pipelines tend to flatten out. The result, in Block’s demos, is a voice that does not just answer questions but converses.
Scam Fears and Education Hopes Split Reactions to Dessa
The flip side of that realism is worry. Fraud watchers fear that voice AI this convincing could fuel scams and deception, making it harder for people to tell a real caller from a synthetic one. Those concerns have followed every leap in voice cloning, but a model that laughs, hesitates and reacts in real time raises the stakes further.
Not everyone sees only risk. Former Binance chief CZ was among those eyeing constructive uses, including children’s education — a setting where a patient, conversational voice could help young learners. Block, for its part, has not disclosed the model’s size, benchmarks or a release date; the research preview is going first to a small group of AI safety researchers.
That cautious rollout suggests Block knows exactly how sensitive this technology is. Dessa may sound like a person, but deciding when — and where — it should be allowed to talk like one is a question the company is still working through.
Source: CoinDesk — read the full story there.
