Block releases speech-only AI trained without text
The model learns language exclusively from human audio and runs locally on M4 Max MacBooks.
By Tomas Rahman ·

Block, the payment company led by Jack Dorsey, announced the release of an artificial intelligence model called Dessa on Thursday. The system was trained entirely on recorded human conversation, with no written text used in its development process. The company published audio clips showing the model interacting with real people to demonstrate its capabilities.
Ragavan Thurairatnam, the project lead, shared one sample exchange where the model initially mistakes the speaker for a person named Ronaldo. Other examples on the Dessa website show varied results. In one clip, the conversation about Christopher Nolan films drifts into a discussion of Bryan Singer’s Superman movies. Another features a user describing a surprise trip to Paris, prompting a response described as teary gratitude. A third clip pairs Dessa with OpenAI’s GPT-Live-1 voice model, where the output appears more natural.
The development team describes Dessa as a true speech-to-speech model. They state that everything the system knows came from human speech. The model is designed to run locally on a MacBook Pro M4 Max or better hardware. This local processing requirement distinguishes it from cloud-based voice assistants that rely on remote servers.
Thurairatnam and his team cited three main reasons for building the model: curiosity, improved human-AI understanding, and safety. They compared the training process to how babies learn to talk without seeing written words. The team noted that while this approach is possible, it is very difficult to execute. They argued that current voice assistants flatten speech into text, losing the natural timing and nuances of human interaction.
The safety argument focuses on emotional perception. The team claims that many AI systems remain nearly deaf to the emotion in human voices as they gain more power. They presented a scenario where a user is in serious pain, but a standard voice AI carries on as if nothing is wrong. Dessa is positioned as a step toward fixing this gap by processing audio directly.
Block plans to give early access to the model to a small group of AI safety researchers. The announcement did not include a timeline for a wider public release. The company has not stated whether the model will be integrated into its existing financial products or stand alone as a separate application.






