Block has unveiled Dessa, a research preview of what it calls the most authentic full-duplex speech model ever built. Unlike mainstream voice assistants, Dessa was trained from scratch on nothing but human conversation, with no text and no pretrained models behind it, according to the company's announcement.
The model is the work of a small AI research lab at Block, the fintech company led by Jack Dorsey. The lab was founded by former members of Dessa, the Toronto deep learning startup, and reports directly to Dorsey himself. A waitlist is now open for access.
How Dessa works and what it can do
Block says the core flaw in today's voice AI is its pipeline: speech gets converted into text, and much of what makes speech meaningful is thrown away. Tone, pauses, breathing, sarcasm and emotion do not survive that conversion, which the company argues makes assistants feel stilted, slow and annoying.
Dessa is a true speech-to-speech model. It is full-duplex, meaning it listens and talks at the same time, much like a person does. Because it learned from real conversation, it knows when to cut in and when to backchannel with small sounds such as "yeah" or "got it" to signal it is following along. Inference runs faster than real time, so there is no awkward wait, and it runs locally on a MacBook Pro M4 Max or better.
- Full-duplex speech-to-speech: listens and talks simultaneously
- Trained only on human conversation audio, with no text or pretrained models
- Learned language from hundreds of times fewer words than large language models
