Block's Dessa is a voice AI trained only on human speech, with no text at all
Block's Dessa learned to talk from conversation audio alone, listens while it speaks, and runs on a MacBook Pro M4 Max. It's a research preview.
The skim
- Block, Jack Dorsey's payments company, has unveiled Dessa, a voice AI model it says was trained "from scratch on nothing but human conversation," with no text and no pretrained models.
- It's full-duplex, so it listens and talks at the same time, and Block says it runs locally on a MacBook Pro with an M4 Max chip or better.
- You can't use it yet. It's a research preview with a waitlist, and a small group of AI safety researchers gets early access first.
What Block announced
Most voice assistants turn your speech into text, run it through a text model and read the answer back. Dessa skips the text entirely. "Audio in, audio out," Block's Dessa page says. "Dessa is a true speech-to-speech model, and everything it knows came from human speech."
Because it learned from real conversations, Block says Dessa knows when to cut in and when to backchannel with a "yeah!" or "got it!", and that it hears and produces emotion, sarcasm and laughter that a text pipeline throws away. Block says inference runs faster than real time.
The lab also makes two efficiency claims: Dessa learned language from "hundreds of times fewer words" than text-based models need, and it was trained on a single Nvidia GB200 NVL72 system.
How it sounds
Block posted live recordings of people talking to Dessa, plus one of Dessa chatting with OpenAI's GPT-Live-1 voice model. The Verge found the conversations mostly natural but sometimes "awkward and uncanny." In one clip, the model starts off thinking the speaker is someone named "Ronaldo."
Who's behind it
Dessa comes from a small AI research lab inside Block, founded by former members of Dessa, the Toronto deep-learning startup, and reporting directly to Dorsey. The team gives three reasons for building it: curiosity, better understanding between people and AI, and safety. "Imagine being in serious pain while a voice AI has no idea and just keeps going," the team writes. "That's the safety problem."
The fine print
Everything so far is Block's own claim. It hasn't published a model size, benchmarks, latency figures, supported languages or a technical paper, and there's no API, download or price. The demo clips are chosen by Block, so they show what Dessa can do, not how often it does it.
What's next
Block is taking early-access applications from AI safety researchers on the Dessa page and wants outside eyes on a model "that learned to talk from sound alone." A wider release hasn't been announced.
