Jack Dorsey’s Block Introduces Dessa, an Emotionally Expressive Voice AI That Runs Locally
aiinsider ·
Jack Dorsey’s company, Block, has launched Dessa, a new voice AI model developed by its research lab that operates locally on consumer hardware. The model is capable of producing emotionally expressive speech, including laughter, pauses, and crying, without relying on external text-to-speech conversions. Unlike many competitors that use separate large language models for knowledge, Dessa derives its conversational understanding directly from its training data. The system functions as a full-duplex model, allowing it to speak and listen simultaneously, similar to OpenAI’s GPT-Live. Block claims this approach avoids the information loss associated with converting speech to text, resulting in a more natural user experience. Although trained on human speech using a single NVIDIA GPU rack, the model is not yet publicly available but is being tested by safety researchers. It can run on devices such as the MacBook M4 Max, potentially helping Block power more voice experiences at lower costs.
aiinsider