Block Unveils Dessa, a Human-Sounding Voice AI That Runs Locally on Consumer Hardware
Block says the model processes speech directly, avoiding transcription pipelines that can erase emotional nuance, and speaks and listens simultaneously. Trained on human speech with one NVIDIA GPU, it runs locally on a MacBook M4 Max and can laugh, pause, or sound tired.
Jack Dorsey’s fintech company Block released a new voice model called Dessa in October 2026, developed out of its research lab. The model is designed to produce speech that sounds distinctly human, including expressions such as laughs, pauses, and emotional shifts like sounding tired or breaking down mid-sentence.
Dessa is a full-duplex model, meaning it can speak and listen simultaneously — a capability comparable to OpenAI’s GPT-Live model. Unlike most voice AI systems, which convert speech to text before processing it, Dessa operates entirely on speech data. Block says this speech-first approach preserves emotional nuance that text-based pipelines discard. The model was trained on human speech using a single NVIDIA GPU.
“Interacting with current voice AI can just plain suck. It turns your voice into text and throws away an incredible amount of information text alone can’t carry. The most powerful AIs in the world can’t hear how you feel,” Block wrote in a blog post accompanying the release.
Block published several demonstration conversations showing Dessa adopting different voices, cadences, and emotional tones. In one example, the model sounded audibly tired while discussing fatigue; in another, a team member played a guessing game with the assistant. The company claims emotion is built into the model itself, though it has not clarified whether any tagging system steers those emotional outputs.
Dessa can run locally on a MacBook M4 Max or comparable consumer-grade hardware. There is currently no public access to the model. Block has issued a call for safety researchers to test it, likely to evaluate risks around voice-based deception.
Block already uses voice AI through its Square platform to handle ordering for small businesses. A proprietary on-device voice model could allow the company to expand those capabilities at lower cost across both business and consumer applications.