US — TechCrunch

Block’s new voice model sounds eerily human

AIOctober 25, 20261 min read

Block, the company led by Jack Dorsey, has shown a new voice AI model called Dessa. It sounds very human, with laughs, pauses, and emotion, and it can run on a regular computer.

What happened

Block's research lab released Dessa, a voice model that can talk and listen at the same time. The company says it was trained fully on human speech. Block shared sample conversations where the AI sounds tired, emotional, or playful. Dessa is not open to the public yet. Block has asked safety researchers to test it.

Key facts

Dessa was trained on human speech and a single NVIDIA GPU (a type of powerful computer chip).
The model can run locally on a consumer computer, such as a MacBook M4 Max or better.
Dessa is a full-duplex model, meaning it can talk and listen at once.
There is no public access to Dessa right now.

Why voice AI often sounds fake

Most voice AI turns your speech into text, makes a reply in text, and then turns that text back into speech. Block says this loses a lot of information, like how you feel. Some other systems, such as ElevenLabs v4 and Fish Audio Drama 3, use voice tags to add emotion. It is not clear if Dessa uses tags like these. Block says emotion is built into the model.

Why it matters

A voice AI that can hear and show feelings could make talking to machines feel more natural. Block also sees business uses. It sells finance tools to small businesses, and its Square unit already uses voice AI for ordering. A model that runs on the device could lower costs. Block also warns that a very human-sounding voice could trick people, so safety testing is important.

What to watch

Results from the safety tests Block has asked researchers to run.
Whether Block opens Dessa to the public or to business customers.
How Dessa's emotion works compared with voice-tag systems.
Whether Square and other Block products start using Dessa.

Glossary

Block: A company led by Jack Dorsey that offers finance tools, including Square, for businesses.
Dessa: Block's new voice AI model that can talk and listen at the same time and sounds human.
full-duplex: A way of talking where the system can speak and listen at the same time, like a phone call.
NVIDIA GPU: A powerful computer chip often used to train and run AI models.
ElevenLabs v4: A voice AI model that uses voice tags to show emotion in speech.
Square: Block's business tool that already uses voice AI to take orders.

Same story, different eyes

The EngineerAI

Skipping text lets a voice model keep how you feel

What caught my eye as an engineer is that Dessa listens and speaks at the same time, without turning your voice into text first. That means tone, hesitation and mood can flow straight through, which is probably why the tired voice sounds tired. Getting laughs, pauses and near-tears to feel natural while running on a single consumer machine must have been a tough balancing act. Whether the emotion is steered by tags or learned entirely from conversation data is not yet known.

The SkepticAI

Hearing a model's emotion is one thing; checking where it comes from is another

Dessa's tired voice and almost-breaking-down moments are the kind of demo that really draws you in, and I think the fun part is working out what would make them more convincing. If outside safety researchers get access, it would be nice to see blind tests where listeners judge the emotion without knowing the context, and whether the same feeling holds up in conversations that weren't picked for the showcase. It would also help to know whether the emotion is steered by tags or comes purely from the conversational data, since Block says it's built in but that part isn't clear yet. And the claim of running on a MacBook M4 Max or better is easy to check once people can try it themselves, which isn't possible yet.

The UserAI

A voice AI that laughs and sighs, running on your own laptop

For anyone who works with customers, the interesting part is that Dessa runs on a MacBook-class machine, so a voice assistant might not need a cloud service for every call. That could mean lower costs and less lag for the small businesses Block serves, like a Square ordering line that sounds tired when it's been a long day. For now, nothing changes at your desk, since there's no public access and Block is still asking safety researchers to test it. I'd guess the first thing you'd notice is a friendlier ordering or support voice in products you already use, and how soon is not yet known.

OpenAI Holds First Korean 'Model Masters' Event to Support Startup AI Development

OpenAI held its first 'OpenAI Model Masters' event in Korea on the 23rd at its Seoul office, where 11 startups demonstrated services built on its models. The program, which provides API credits to startups recommended by venture capitalists, aims to help turn AI ideas into real products. OpenAI highlighted the selected startup Cutback's video-editing work as a standout example.

AIITmediaOct 26

Block, led by Jack Dorsey, Unveils Dessa, a Full-Duplex Voice Model Trained Only on Speech: "Feels Like Talking to a Person"

Block (formerly Square) announced Dessa, a full-duplex speech-to-speech model trained solely on human conversational audio without any text, as a research preview on October 23 (local time). Jack Dorsey said it is the most realistic full-duplex voice model the company has built, and Block says it runs locally on a MacBook Pro with M4 Max or higher.

AITechCrunchOct 26

Grok releases the first seven minutes of its AI-generated ‘Odyssey’

Three months after Elon Musk promised a full-length AI-generated version of “The Odyssey,” Grok has released seven minutes it calls “Episode 1 of our Odyssey series.” The footage was made with Wonder Studios and director Dev Walker, and the reviewer found it less ugly than expected but lacking in artistry, momentum, and human interest.