A speech model that can listen and respond at once
Block has introduced Dessa, a voice AI model built to make spoken interactions feel more natural. The company says it was trained entirely on human speech and can run locally on consumer-grade hardware, including a MacBook M4 Max or a more powerful computer, with a single NVIDIA GPU.
Dessa uses a full-duplex approach, allowing it to listen and speak at the same time. Unlike systems that convert speech into text before generating a response, the model is designed to retain vocal cues such as pauses, cadence and emotion. Block says its conversational knowledge comes from its speech-training data.
Block describes the model as capable of varied voices and expressive delivery, including changes in tone that reflect the conversation. The company has not yet opened public access, and is inviting safety researchers to assess how voice agents could be used responsibly.
For Block, which serves businesses through products including Square, an on-device voice model could support new customer and business services while reducing reliance on cloud processing. If its expressive capabilities prove useful and trustworthy, Dessa could help make voice a more intuitive interface for everyday technology.
