Block (the company behind Square and Cash App) unveiled Dessa, a new voice model trained exclusively on human conversations. It was shown off by Twitter co-founder and Block CEO Jack Dorsey.
How it works
Unlike most voice assistants, Dessa doesn't convert speech to text and back to audio. It processes and generates sound directly — which means it can listen and respond at the same time.
The model can:
- laugh;
- pause;
- interrupt;
- shift intonation;
- react to emotions;
- sound like it's about to cry.
The demo
In the demo, Dessa doesn't behave like a typical chatbot: it laughs mid-sentence, says "mm-hmm," reacts to the person it's talking to, and changes tone in real time during the conversation. Dorsey posted the demo video himself.
Why it exists
Block's idea is to make talking to AI feel less like talking to a machine and more like an ordinary conversation with a person. For now, Dessa is only available to a limited group of AI safety researchers.
What's unsettling
Its unnervingly natural delivery already raises the big question: how will we know who's actually on the other end of the phone? This kind of technology could make voice assistants a lot more pleasant — and voice scam calls a lot more convincing.
It looks like voice AI has finally gotten realistic enough that this isn't just an impressive demo anymore.