Block launches Dessa, a voice AI model trained only on human speech

Block, the company co-founded by Jack Dorsey, has released a new artificial intelligence system called Dessa. The company describes the model as being trained from scratch using nothing but human conversation. To demonstrate the software, Block published several audio clips of the model interacting with people. Some of these exchanges appear natural, while others sound awkward or uncanny to listeners.

Ragavan Thurairatnam, who leads the Dessa project, shared one sample recording. In that specific exchange, the model initially mistakes the speaker for a person named Ronaldo. The Dessa website hosts additional examples of the software in action. One clip features a discussion about Christopher Nolan films that shifts into a tangent about Bryan Singer’s Superman movies. Another recording shows a user telling the model about a surprise trip to Paris. The website notes that Dessa responds with what sounds like teary gratitude in that instance.

The team behind the project states that Dessa is a true speech-to-speech model. They claim that everything the system knows came directly from human speech. The developers note that the model can run locally on a MacBook Pro M4 Max or better hardware. They also paired Dessa with OpenAI’s GPT-Live-1 voice model in one clip, where the interaction sounded more natural than in the standalone examples.

The developers list three primary motivations for creating the model: curiosity, better understanding between humans and AI, and safety. Regarding curiosity, the team compares the process to how babies learn to talk without ever seeing written words. They state that they wanted to see if a model could achieve the same result and note that while it is possible, it is very hard to execute.

The team argues that current voice assistants waste the potential of speech. They believe speech is the most natural interface for any intelligence, but say today’s voice AI flattens what people say into text and cannot keep up with human timing. On the safety front, the developers argue that AI systems are gaining power while remaining nearly deaf to the emotion in human voices. They describe a scenario where someone is in serious pain, but the voice AI does not realize it and continues the conversation as usual. They describe Dessa as a step toward fixing that issue.

Block plans to provide early access to the model for a small group of AI safety researchers. The announcement did not include a schedule for a wider public release.