Block has introduced Dessa, a speech AI model designed to make conversations with AI sound more natural by capturing emotion, pauses and human speech patterns. It can listen and respond simultaneously and is designed to run locally on consumer hardware.
According to Block’s research team, Dessa was trained entirely on human speech using a single NVIDIA GPU rack. Unlike conventional systems that convert speech into text before generating a response, it processes spoken interactions directly.
Follow THE FUTURE on LinkedIn, Facebook, Instagram, X and Telegram
Real-Time Speech And Emotional Expression
Dessa uses a full-duplex architecture, allowing it to listen and speak at the same time rather than waiting for a person to finish. Block says this approach helps preserve vocal cues that can be lost when speech is converted into text.
The company also says Dessa draws knowledge from conversational data within the model itself, rather than relying on a separate large language model. Its full capabilities and limitations have not yet been disclosed.
“We believe voice is the most natural way to interact with intelligence,” Block CEO Jack Dorsey wrote in a post announcing Dessa, highlighting its ability to convey emotion, nuance and natural cadence.
Demonstrations shared by Block researchers show the model adapting its tone to different situations. However, these examples do not yet establish how consistently it can interpret emotion in real-world conversations.
Local Processing And Safety
Block says Dessa can run on consumer hardware, including a MacBook with an M4 Max chip or better. Local processing could reduce latency, cloud computing costs and reliance on external infrastructure.
The model is not yet publicly available. Block has invited safety researchers to test scenarios in which a convincing AI voice could mislead someone into believing they are speaking with a human.
Potential Applications
Block’s ecosystem includes products for consumers and small businesses, creating potential applications in commerce and customer service. Its Square platform already supports AI-powered voice ordering.
Dessa could eventually support additional voice-enabled services while reducing reliance on cloud processing. However, its commercial potential will depend on reliability, safety and performance on everyday devices.







