Block Trains Its New Voice AI Only on Spoken Words

Block announced an artificial intelligence model called Dessa on Thursday that it says was built "from scratch on nothing but human conversation." The company, led by Jack Dorsey, published recordings of people talking with the system to show how it performs.

Ragavan Thurairatnam, who leads the Dessa project, posted one exchange in which the model sounds largely natural but first takes the speaker for someone named "Ronaldo." Other clips on the Dessa site include a chat about Christopher Nolan's films that drifts into Bryan Singer's Superman movies, and a person telling the model about a surprise trip to Paris, after which the site says Dessa "then answers with what sounds like teary gratitude." A third clip puts Dessa against OpenAI's GPT-Live-1 voice model, where Dessa sounds more natural.

The Dessa team calls it a "true speech-to-speech model, and everything it knows came from human speech," and says it runs locally on a MacBook Pro M4 Max "or better." The developers list three reasons for building it: "curiosity (aka fun), better understanding between humans and AI, and safety." On the first, they compare the system to infants: "Babies learn to talk without ever seeing a word written down. We wanted to see if a model could too. It can. It's just very hard."

The team also argues that today's voice assistants flatten speech into text and cannot match human timing, and that AI systems remain "nearly deaf to the emotion in our voices" even as they grow more capable. They describe Dessa as "a step toward fixing" that gap, citing a case of someone in serious pain whose voice AI keeps talking without noticing.

Block plans to give "a small group of AI safety researchers" early access to Dessa. The announcement did not set a wider release schedule.