Block unveils Dessa, a voice AI trained only on human speech that runs on a laptop

What is Dessa?

Dessa is a voice AI from a small research lab at Block, the company behind Square and Cash App. The lab reports to Jack Dorsey, Block's cofounder and chairman, and was founded by former members of Dessa, the Toronto deep learning startup.

Block calls it a true speech-to-speech model: audio goes in and audio comes out, with no text step in between. Block says that makes it the first AI you can have a realistic live conversation with that was trained only on human speech.

How was Dessa trained?

Block says it trained Dessa from scratch on nothing but human conversation: sound, with no text and no pretrained models. It says the model learned to hold a real-time conversation simply by listening.

Block also says Dessa learned language from hundreds of times fewer words than large language models need, and that it was trained on a single Nvidia GB200 NVL72 system. These are Block's own claims.

What can Dessa do?

Block says Dessa is full-duplex: it listens and talks at the same time, like a person. It says the model knows when to cut in and when to say "yeah" or "got it" to show it is following, and that it hears and speaks emotion, sarcasm and laughter, which a text step would lose.

Block shared recordings of live conversations with it, including a life-advice chat and a conversation with GPT-Live-1, a third-party AI model. Block says nobody steers Dessa in the clips and that subtitles were added afterwards.

Can I run Dessa on my own computer?

Block says Dessa runs faster than real time, locally, on a MacBook Pro M4 Max or better. But Block's page says launch links are coming soon, so there is nothing to download yet.

Block is giving a small group of AI safety researchers early access first, to find risks it has not anticipated. People can request access on its page with an institutional email.

A model that learned to talk from sound alone is new territory, and we want outside eyes on it.

Ragavan Thurairatnam, Dessa lead at Block

Why does it matter?

Most voice AI today turns speech into text, thinks in text, then turns the answer back into speech. Block argues that throws away emotion and timing. If Dessa works as Block describes, voice AI could feel much more like talking to a person.

That also raises questions about how convincing a synthetic voice should be. The samples are Block's own picks, so outside tests will show how it does in everyday use.

What's next?

Early access for safety researchers comes first. Block has not given a public release date.