Brief
Block launches Dessa, a full‑duplex voice AI that mimics human conversation
Block says its speech‑to‑speech model Dessa sounds uncannily human and runs locally on high‑end MacBooks.
By Felo News Desk · Published
Block, the company formerly known as Square, announced a new AI model called Dessa that creates speech directly from speech, according to a post on the firm’s X account. The model, described as "trained from scratch on nothing but human conversation," is designed to sound like a person, complete with emotion, nuance, pauses and a natural cadence.
What happened
Dessa uses a full‑duplex architecture that lets it listen and speak at the same time, avoiding the abrupt silences typical of many voice assistants. Block released several audio demos, including a video where a Block intern plays a guessing game with the model, and another where a differently voiced version of Dessa answers in an African‑American male cadence, saying, "man, I’m doing good, breh. I ain’t gonna even lie. I’m tired." The model claims to run locally on any MacBook Pro M4 Max or better.
What the reports add
The Verge reported that Dessa processes and generates audio directly using speech‑to‑speech architecture rather than converting input speech to text first. The same outlet noted that the duplex design lets the model respond without the "subservient silences" other AIs exhibit when a user interrupts.
What was said
Ragavan Thurairatnam, the team lead for Dessa, posted the demo videos on X, describing the model’s ability to guess objects on the first try and to continue guessing after being told it had won. Block’s announcement post claimed that talking with Dessa feels like talking with a person.
How it came about
Block, which rebranded from Square in 2021, has a history of launching consumer‑facing tech products such as Cash App and Afterpay. The company says it has already granted a small group of AI safety researchers early access to Dessa, positioning the model as a testbed for future voice‑AI developments.
Key facts
- Block released an AI model named Dessa that generates speech directly from speech. (gizmodo.com)
- Dessa uses a full‑duplex architecture allowing it to listen and speak simultaneously. (gizmodo.com)
- Demo videos show Dessa guessing objects and speaking in varied human‑like cadences. (gizmodo.com)
- Block says the model can run locally on any MacBook Pro M4 Max or better. (gizmodo.com)
- A small group of AI safety researchers have been given early access to Dessa. (gizmodo.com)
Sources
- [1] gizmodo.com — originally reported as “Check Out the Many Uncanny, Human-ish Voices From Block's New AI Model”
