Block, the payments company led by Jack Dorsey, has unveiled Dessa, a voice AI model trained entirely on human speech that can listen and talk at the same time, TechCrunch reported.

What's different

  • No text: Block says it trained Dessa from scratch on human conversation, with "no text, no pretrained models." Most voice systems convert speech to text and back, which Block says throws away emotion and tone.
  • Full duplex: Dessa can talk and listen simultaneously, knowing when to cut in and when to say "yeah!" or "got it!" to show it's following, according to Block.
  • Runs locally: The model runs on a MacBook Pro M4 Max or better, and inference is faster than real time, Block said. TechCrunch reported it was trained on a single NVIDIA GPU rack.

"Interacting with current voice AI can just plain suck," Block wrote. "The most powerful AIs in the world can't hear how you feel."

Not public yet

There is no public access to Dessa yet. Block has put out a call for safety researchers to test it, TechCrunch reported. Square already uses voice AI to handle ordering, and an on-device model could power more business and consumer features at lower cost.