Block’s new voice model sounds eerily human

Jack Dorsey’s Block has unveiled Dessa, a new full-duplex speech-to-speech voice model developed by its research lab that runs locally on consumer hardware such as a MacBook M4 Max or better, trained entirely on human speech using a single NVIDIA GPU. Unlike many existing voice AI systems that convert speech to text and lose emotional nuance, Dessa is designed to talk and listen simultaneously and has emotion built into the model, producing natural laughs, pauses, and cries. Block claims Dessa is the most authentic full-duplex speech model ever built, though it remains unclear whether emotions are steered by tags similar to ElevenLabs v4 or Fish Audio Drama 3. The company has released examples of conversations showing varied voices and emotional expressions, and argues that current voice AI often provides a worse and potentially dangerous experience by discarding information that text cannot carry. Dessa is not yet publicly available; Block has called for safety researchers to test it, likely to examine risks such as voice agents tricking people. The model could eventually power more business and consumer experiences at lower cost, building on Square’s existing use of voice AI for ordering.

(Source：TechCrunch)