THE CRUNCH
Block, Jack Dorsey's payments company, has unveiled Dessa, a voice model from its research lab that runs locally on consumer hardware and produces strikingly human speech, complete with laughs, pauses and cries. The company says the model was trained fully on human speech using a single NVIDIA GPU, and that it can run on a consumer-grade computer such as a MacBook M4 Max or better.
Dessa is a full-duplex model, meaning it can talk and listen at the same time, an approach comparable to the GPT-Live model OpenAI released earlier this year. Where many speech models are paired with a large language model for knowledge retrieval, Block says Dessa's knowledge comes from its conversational data.
Block argues that current voice AI can sound unnatural because text sits in the pipeline driving speech generation, discarding the emotional information that text alone cannot carry. Rivals such as ElevenLabs v4 and Fish Audio Drama 3 use voice tags to steer emotion, and Block claims emotion is built into Dessa itself, though it is not clear whether tags play a role behind the scenes. Researchers posted demo conversations showing the model shifting voices, cadences and moods, including one where it sounded tired while saying it was tired from work.
There is no public access to Dessa at the moment. Block has instead put out a call for safety researchers to test the model, likely to probe cases where a voice agent could trick the person on the other end.