Jack Dorsey’s Block Unveils Dessa, a Human-Like Speech AI Model
- Jack Dorsey’s Block unveiled an experimental artificial intelligence voice model named Dessa on Oct.
- Unlike conventional voice assistants such as ChatGPT or Google’s Gemini that convert speech into text before processing a response, Dessa operates on a full-duplex architecture.
- Block stated that the model was trained from scratch on recordings of human speech without any underlying text corpora or pretrained large language models.
Jack Dorsey’s Block unveiled an experimental artificial intelligence voice model named Dessa on Oct. 23, 2026, launching a speech-to-speech system trained entirely on human conversation rather than text, Mashable reported. The release immediately drew sharp division online, with observers praising its natural cadence while warning of potential security risks.
Full-Duplex Architecture and Speech-Only Training
Unlike conventional voice assistants such as ChatGPT or Google’s Gemini that convert speech into text before processing a response, Dessa operates on a full-duplex architecture. According to BigGo Finance, the system listens and talks simultaneously, allowing it to interject, backchannel, and react without waiting for another speaker to finish.
Block stated that the model was trained from scratch on recordings of human speech without any underlying text corpora or pretrained large language models. This direct audio processing pipeline enables the model to capture acoustic nuances like laughter, sighs, sarcasm, and breaks in a speaker’s voice, as detailed by BigGo Finance. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence,
the company said.
Internet Reactions and Fraud Concerns
While some users praised the human-like nuance and conversational flow, others raised alarms regarding the potential for advanced fraudulent phone calls. Some called it “scary” and “way more dramatic” and sound like “illiterate teenager.”
Critics pointed out that voices indistinguishable from actual humans could exacerbate telephone scams targeting vulnerable populations.
Get ready for a new era of scam calls!
Unnamed social media user cited by The News International
Some observers focused on how far the model’s cadence has moved from the clipped, deferential tone of existing assistants, and Binance founder Changpeng Zhao pointed to possible classroom uses, writing that teaching children with current AI voices had been a challenge.
Early Access and Hardware Requirements
Block’s research team, led by Ragavan Thurairatnam and featuring former members of the Toronto deep-learning startup Dessa that Square acquired in 2020, has granted early access to a small group of artificial intelligence safety researchers, Gizmodo reported. The unit operates under Block chairman and chief executive Jack Dorsey.
The company has not disclosed model parameter weights, benchmark results, or a formal public release date. However, technical demonstrations indicate the model can run locally on consumer hardware, requiring a MacBook Pro equipped with an M4 Max chip or better, according to BigGo Finance.
Background and Technical Demonstrations
If interacting with Siri has been the extent of your experience talking
to AI, you’re in for a rude awakening.