Dessa: the AI that listens | Telephonist Updates

Block has unveiled Dessa, a voice AI trained only on recordings of people talking, with no text at all. It listens while it talks, laughs mid-sentence and hears tone. Here are six of Block’s recordings.

Dessa comes from a small research lab at Block, the company behind Square and Cash App. Block says the lab trained it from scratch on human conversation alone: no text and no pretrained models. It learned to hold a conversation by listening.

Most voice AI turns what you say into text, works out a reply, then reads it back. Tone, pauses, a sigh or a laugh get lost on the way. Dessa never makes that conversion: it hears sound and answers in sound. It also listens while it talks, so it can laugh mid-sentence, finish your words or say “yeah” while you’re still speaking. Block says it runs on a MacBook Pro with an M4 Max chip or better.

Why it matters on the phone

Telephonist lives on phone calls, where how something is said often matters as much as the words: a caller who sounds worried, rushed or upset. A model that hears tone directly could help AI on the phone respond much more like a person does. Block makes the same point about safety: an AI that can’t hear distress in a voice may simply carry on.

The more human AI voices sound, the more it matters to know when you’re talking to one. That’s why every call our Telephonist makes says it’s an AI, and why any business can check a call from us.

Six recordings

Each is a live conversation between a person and Dessa, recorded and subtitled by Block. Block says nobody steers it: Dessa picks what to say, when to say it and how. Tap a label to jump to that moment.