Block, founded by Twitter creator Jack Dorsey, has unveiled a new artificial intelligence model named Dessa. The system is distinguished by its training process, which was conducted entirely from scratch using only human speech data. Unlike many existing voice models, Dessa operates as a full-duplex system, allowing it to listen and speak simultaneously during interactions.
The model is designed to function as a speech-to-speech system, deriving its knowledge base exclusively from spoken language. According to the developers, the software can run locally on hardware such as an Apple MacBook Pro equipped with an M4 Max processor or more powerful computing devices. The project aims to improve emotional recognition and human-AI interaction.
Block stated that the development of Dessa was driven by goals related to curiosity, enhanced human understanding, and safety. The company highlighted that current voice AI systems often rely on text conversion, which limits their ability to process emotional nuances in real-time. Dessa is intended to address these limitations by responding directly to vocal cues.
At this stage, access to the Dessa model is restricted. The company has announced that early access will be granted to a select group of researchers specializing in AI safety. No timeline for a broader public release has been provided by the developers at this time.
