Block’s New Dessa AI Voice Sounds So Human It’s Making People Nervous
Quick Reads
- Block released a voice model called Dessa from its research lab. It runs locally and has human-like expressions such as laughs, pauses, and cries.
- Dessa trained entirely on human speech rather than text, using a single Nvidia GB200 NVL72 system.
- It is “full-duplex,” meaning it can talk and listen at the same time.
- On a consumer-grade computer such as a MacBook M4 Max or better.
- Can you try it? No. There is no public access yet, and Block has called for safety researchers to test it.
- The catch: Block hasn’t disclosed the model’s size, benchmark results, or a release date, and the audio samples are company demos, not independent tests.
The Metaphor: From Reading a Script to Hearing a Heartbeat
Most voice AI today works like an actor reading a script. Your voice is turned into text, a language model writes a reply, and a speech engine reads it aloud. The words are right, but the performance is flat.
Dessa is closer to a jazz musician improvising with you. It listens while it plays, so it can react to your timing, your laugh, and your interruptions as they happen. Block says that’s the point: turning speech into text throws away information that text alone can’t carry.
What Block Actually Built
Typically, speech models are paired with a large language model for knowledge retrieval. Block says Dessa’s knowledge comes from the conversational data it learned from. In other words, it is a speech-first model, not a text model wearing a voice.
It processes audio directly, which lets it handle interruptions, laughter, and changes in tone mid-conversation. That is a big departure from the usual pipeline.
Block argues that many current models sound unnatural because text sits in the pipeline driving the speech. Tools like ElevenLabs v4 and Fish Audio Drama 3 use voice tags to steer emotion. Block says emotion is built into Dessa, though it isn’t clear whether any kind of tagging steers it.
What the Demos Show
In one demo, the bot said it was tired from work, and it sounded tired. In another, it talked emotionally about someone making it feel comfortable and almost seemed to break down. A Block team member also played a guessing game with it, and it guessed the first answer right away.
In another clip, it laughs in the middle of a sentence. Block says the conversations were recorded live with nobody directing the model’s responses.
The Public Reaction
The awe camp: One X user said he had never heard anything artificial that sounded so human, and another called it “straight up sci-fi stuff.”
Binance founder Changpeng “CZ” Zhao said it could help his children’s education project, since teaching kids with today’s AI voices was a challenge.
Filmmaker Justine Bateman asked why AI voices need to sound human at all, suggesting it could be used to trick people into lowering their guard.
Journalist Shiv Aroor warned that scam calls are about to reach pandemic levels.
Why Block, and Why It Matters
Block sells financial tools to small businesses, and Square already uses voice AI to handle ordering. A voice model that runs on-device could power more business and consumer experiences at lower cost.
Block also operates Cash App, the bitcoin wallet Bitkey, and the mining hardware unit Proto. Its AI lab reports directly to Dorsey and includes former staff of the Toronto AI startup Dessa, which Square acquired in 2020.
For everyday people, the practical stakes are: Better customer service: AI that can hear frustration and respond with care.
Cheaper automation: a model small enough to run on a laptop costs less to deploy than one in a data center.
Harder-to-spot fraud: if a voice can laugh, hesitate, and sound distressed, the old “it sounds robotic” test stops working.
What We Still Don’t Know
How big the model is and how it performs on independent benchmarks, when, or whether, the public will get access. Whether it can be misused to imitate specific real people and what the safety researchers, who are getting early access first, will find. Block says they will assess risks the company may have overlooked.
Bottom Line
Dessa is not available to the public, and its best moments come from Block’s own demos. But if they hold up, voice AI is moving from sounding correct to sounding alive. That is exciting for education and customer service, and a warning for anyone who answers unknown phone calls.
Stay-safe tip for readers: Don’t trust a voice alone for money requests. Hang up and call back on a number you already know.
