Block Unveils Dessa: A Human-Like Full-Duplex Voice AI Model
Block has announced 'Dessa,' a new full-duplex speech-to-speech model developed by its research lab. Unlike traditional models that convert voice to text before processing, Dessa is trained directly on human speech, allowing it to capture nuances such as laughter, pauses, and emotional shifts. The model is designed to run locally on consumer-grade hardware, such as a MacBook M4 Max, significantly reducing the need for massive cloud computing resources. While there is currently no public access to Dessa, Block has invited safety researchers to evaluate potential risks, including the possibility of voice agents deceiving users. For Block, which operates Square and other financial services, this technology could potentially lower costs and enhance consumer and business experiences by integrating highly naturalistic AI interactions into its existing ecosystem.
Topics: Artificial Intelligence Product Innovation Speech Synthesis Voice AI