Models · 2026-10-24
Block shows Dessa, a voice model that sounds almost human
Jack Dorsey's company Block has built Dessa, a voice AI model that can laugh, pause, and speak with emotion, but it is not yet publicly available to try.
What this model is
Block, the company led by Twitter co-founder Jack Dorsey, has released a new voice model called Dessa. It was created in the company's research lab. The model can run locally, meaning on an ordinary computer without a constant connection to the cloud. Block says it was trained entirely on human speech, using a single rack of Nvidia GPUs.
The key difference is how it talks. It can sound tired, upset, or cheerful, and it can laugh, pause, and even cry. The company claims emotion is built directly into the model, while many other systems rely on special emotion tags to control how they sound.
How it differs from typical voice assistants
Usually, voice AI first turns speech into text, processes that text, and then generates a spoken reply. Block says this approach loses a lot of information, because text cannot convey how a person feels. That is why, according to the company, current systems often sound unnatural.
Dessa is a full-duplex model, meaning it can talk and listen at the same time, the way people do in real conversations. OpenAI has shown a similar approach with its GPT-Live model. Block says Dessa's knowledge comes from the conversational data it was trained on.
Can you try it
There is no public access to Dessa yet. Block has invited safety researchers to test the model, mainly to check whether a voice AI could trick people on the other end of a call. The company says the model can run on a consumer computer, such as a MacBook M4 Max or better.
For Block, this is also a business matter. The company works in finance and sells its services to small businesses. Its Square service already uses voice AI to take orders. A voice model that runs on the device itself could make it cheaper to power more business and consumer features.
Useful from the source
- Block's Dessa blog post — Block's official page describing the Dessa voice model.
- Fish Audio Drama 3 — A voice model cited as an example of systems that control emotion with tags.
- Square voice ordering help article — Square's guide to taking orders with AI-powered voice ordering.
Sources: 🔗 TechCrunch — AI