Block's New AI Model Dessa Learns Language Without Reading a Single Word
The system, trained exclusively on human audio, runs locally on high-end Mac hardware and aims to capture vocal emotion that text-based models miss.
By Alina Carter ·
Block, the company founded by Jack Dorsey, has released a new artificial intelligence model named Dessa. The system was built to process and generate speech without converting audio to text or text to audio. Block stated that the model was trained from scratch on nothing but human conversation. This approach removes written language entirely from the training dataset.
The company published several audio recordings to demonstrate the model's capabilities. Ragavan Thurairatnam, who leads the Dessa project, shared one example where the interaction appears mostly natural. In that specific clip, the model initially mistakes the speaker for a person named Ronaldo. Other samples on the Dessa website show the model discussing Christopher Nolan films before shifting focus to Bryan Singer's Superman movies. In another recording, a user describes a surprise trip to Paris, and the model responds with what the site describes as teary gratitude.
Block described Dessa as a true speech-to-speech model. The team said everything the system knows came directly from human speech. The developers noted that the model can run locally on a MacBook Pro M4 Max or better hardware. This local execution allows the system to operate without sending data to external servers for every interaction.
Thurairatnam and his team listed three main motivations for the project. They cited curiosity, a desire for better understanding between humans and AI, and safety. The team compared the training process to how babies learn to talk. They noted that infants learn language without seeing written words. The developers said they wanted to see if a model could achieve the same result, acknowledging that the task is very difficult.
The team argued that current voice assistants limit the potential of speech. They stated that speech is the most natural interface for intelligence. However, they claimed that today's voice AI systems flatten what people say into text and fail to keep up with human timing. The developers believe this conversion process loses critical data about how people speak.
Block also highlighted safety as a key reason for building Dessa. The team said AI systems are gaining power while remaining nearly deaf to the emotion in human voices. They offered a scenario where a person in serious pain speaks to a voice AI that does not detect the distress. Block described Dessa as a step toward fixing that disconnect. The company plans to provide early access to a small group of AI safety researchers. The announcement did not include a schedule for a wider release.
A separate clip pairs Dessa with OpenAI's GPT-Live-1 voice model. In that specific exchange, Dessa sounds more natural than in other samples. The company has not stated whether this integration is a standard feature or a one-time demonstration.
