Jack Dorsey's Block Unveils Local Full-Duplex Voice AI Model 'Dessa'

Jack Dorsey's Block unveiled Dessa, a full-duplex speech-to-speech AI model developed by its research lab that can talk and listen simultaneously while expressing human-like pauses, laughs, and cries. Block stated that Dessa was trained entirely on human speech using a single Nvidia GPU and runs locally on consumer hardware such as an Apple MacBook M4 Max or better, following other recent local voice agent models. Unlike traditional voice assistants that convert audio into text and rely on voice tags like ElevenLabs v4 or Fish Audio Drama 3 to simulate emotion, Block says Dessa processes speech directly without a text pipeline and draws its knowledge from conversational training data. Block has not released public access to Dessa and is currently recruiting safety researchers to test the model before a broader rollout. "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. We think that’s ridiculous, leads to a worse experience, and is potentially dangerous." — Block