The full-duplex model trains on human speech with one NVIDIA GPU and handles emotion without a text pipeline.
Block released Dessa, a voice model from its research lab that runs locally on a consumer machine and produces laughs, pauses, and cries. The company trained the model fully on human speech using a single NVIDIA GPU. Dessa is full-duplex, so it talks and listens at the same time.
Jack Dorsey’s Block built Dessa without a text stage in the pipeline, and the company argues that routing speech through text discards information about how a person feels. Competing systems such as ElevenLabs v4 and Fish Audio Drama 3 rely on voice tags to steer emotion. Block says emotion is built into Dessa, though the company has not detailed how it steers that emotion.
Builders gain a voice model that runs on a MacBook M4 Max or better, which lowers the cost of shipping conversational agents. Block sells finance tools to small businesses, and Square already uses voice AI for ordering. An on-device model lets the company extend voice across more business and consumer experiences without paying per-call cloud fees.
Block has not released Dessa publicly and has called for safety researchers to test it. That review likely targets cases where a voice agent deceives the person on the other end of a call. Access terms, pricing, and a release date for Square and other products remain unannounced.
What matters
- Block released Dessa, a full-duplex voice model that talks and listens at the same time.
- On-device voice AI cuts inference costs for operators building customer-facing agents.
- Block has opened Dessa to safety researchers and still offers no public access to the model.
Why it matters
Block has opened Dessa to safety researchers and still offers no public access to the model.
This GenAI News article was prepared in original wording using reporting and materials published by TechCrunch AI. Source reference: https://techcrunch.com/2026/10/24/blocks-new-voice-model-sounds-eerily-human/.
Drafted by the GenAI News review pipeline.