Jack Dorsey's Block has unveiled a new voice model from its research lab, and it sounds eerily human - complete with laughs, pauses, and cries. The model, called Dessa, runs locally and is trained fully on human speech and a single NVIDIA GPU rack. Because nothing says "authentic human emotion" like a rack of GPUs.

Dessa is a full-duplex model, meaning it can talk and listen at the same time, much like OpenAI's GPT-Live model released earlier this year. Typically, speech models are paired with a large language model for knowledge retrieval, but Block claims Dessa's knowledge stems from the conversational data it was trained on. So it's not just a pretty voice - it's a pretty voice that might actually know things.

Block argues that current models sound unnatural and annoying because they have text in the pipeline driving speech generation. Many models, such as ElevenLabs v4 or Fish Audio Drama 3, use voice tags to steer emotion. Dorsey's company claims emotion is built into Dessa - though whether that emotion is being steered by some kind of tags remains unclear. It's like saying your car has built-in happiness, but not mentioning the smiley-face air freshener.

"Interacting with current voice AI can just plain suck," the company said in a blog post, in a rare moment of corporate honesty. "It turns your voice into text and throws away an incredible amount of information text alone can't carry. The most powerful AIs in the world can't hear how you feel. We think that's ridiculous, leads to a worse experience, and is potentially dangerous." Block has a point: an AI that can't tell you're upset is just a fancy chatbot with a tin ear.

The company's researchers posted multiple examples of conversations with Dessa, where the agent exhibited different voices, cadences, and emotions. In one, the bot said it was tired from work - and sounded tired. Groundbreaking stuff: an AI that can complain about its job just like the rest of us. Another example showed the bot talking about a person making it comfortable while sounding emotional and almost breaking down. Finally, an AI that can have a breakdown without needing a therapist.

Block says Dessa can run locally on a consumer-grade computer such as a MacBook M4 Max or better. However, there's no public access at the moment, and Block has put out a call for safety researchers to test the model - likely to probe cases where a voice agent could trick people on the other end. Because the only thing scarier than an AI that sounds human is an AI that sounds human and is trying to scam you.

Block is in the business of finance and selling solutions to small businesses. Square already uses voice AI to handle ordering, but with its own on-device voice model, Block could power a lot more business and consumer experiences at a lower cost. So next time you're ordering a coffee through Square, don't be surprised if the barista sounds a little too human - and a little too tired.