Block Releases Local Voice Model Called Dessa, Praised for Lifelike Expression

Jack Dorsey’s company Block has introduced a new voice model named Dessa, developed out of its research lab. According to the company, the model is trained entirely on human speech and a single Nvidia GPU rack, and it is capable of running locally on consumer-grade hardware such as a MacBook M4 Max or better.

Block reports that Dessa is a full-duplex model, meaning it can speak and listen simultaneously. The company claims the model's knowledge stems directly from conversational data rather than requiring a paired large language model for knowledge retrieval.

Key Details About Dessa

• Local Execution: Designed to run on consumer-grade computers.
• Full-Duplex Capabilities: Supports simultaneous listening and talking.
• Speech Training: Trained entirely on human speech instead of relying on text pipelines.

Practical Implications and Safety Testing

While Block has shared demonstration examples showcasing various emotions, cadences, and expressive elements like pauses and laughs, public access to the model is currently not available. Instead, the company has issued a call for safety researchers to test the model, addressing potential concerns such as voice agents being used to deceive individuals.

To verify announcements regarding Dessa and check for updates on public availability, interested parties can monitor official company blog posts and communications from Block's research team.