English
Español
Most AI voices still give themselves away. They are smooth, fast, and just robotic enough that you know you are talking to a machine. Block, the company led by Jack Dorsey, says it has built something that closes that gap. On October 24 it unveiled Dessa, a voice model trained only on real human conversations, and early listeners describe it as sounding eerily human.
What makes Dessa different
Almost every voice assistant you have used works in two clumsy steps. It converts your speech to text, processes the text, then converts a text reply back into speech. That round trip is why today’s voices can feel flat and why they miss the feeling behind your words. Dessa skips it. It is a speech-to-speech model that processes audio directly, without translating everything into text in the middle.
Because it works with sound itself, Dessa can pick up the things text throws away: a laugh, a pause, a crack of sadness in someone’s voice, and respond to them in real time. Block trained it exclusively on human conversations rather than synthetic data, which is a big part of why it lands as natural rather than polished. Just as notable, Block says Dessa can run locally on a MacBook M4 Max. A model this capable that does not need a data center to work is unusual, and it changes the privacy conversation entirely.
Why “too human” matters
A voice that can hear how you feel is both the appeal and the unease. On one hand, it points to genuinely better experiences: support calls that sense frustration and adjust, accessibility tools that feel personal, companions that do not sound like a phone tree. On the other, a machine that convincingly mimics human emotion raises hard questions about deception, consent, and how we will tell a real voice from a generated one.
The fact that a recognizable founder like Dorsey is behind it gives the project a face and a point of view. Block is not a sleepy incumbent, and its decision to make Dessa run on consumer hardware suggests a bet that the future of voice AI is private and personal, living on your own device, rather than piped in from the cloud. If that bet is right, the “eerily human” reaction people are having today becomes the baseline expectation tomorrow.
What it means for your business
Voice is one of the last places where customers can still instantly tell they are not talking to a person. A model like Dessa narrows that line to almost nothing, which matters for anyone who talks to customers at scale. Sales calls, support lines, and voice assistants built on this kind of technology could soon read emotion and respond with the warmth that makes people trust a brand.
That cuts both ways. Customers increasingly want to know when they are speaking to an AI, so the businesses that win will be the ones that use emotional, natural voice to serve people better, not to trick them. If your company handles a high volume of calls, it is worth watching where speech-to-speech models land, because the tools that can hear a frustrated customer and adjust in the moment are about to get very good.
Dessa is still an unveiling rather than a product you can buy today, and “sounds human in a demo” is not the same as “sounds human on every call.” But the direction is unmistakable. The text bottleneck that made AI voices feel robotic is falling away, and the people building the next wave of voice tools are aiming straight for something that feels like a real conversation. Whether that delights your customers or unsettles them will come down to how honestly it is used.
