🔍 Read the full analysis: AI Voice Has A Human Rhythm. Block’s Dessa Shows Why That Isn’t Proof on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
Block announced Dessa, a speech-to-speech research model it says was trained on recorded human conversation and can respond with laughter, pauses and interruptions. Its demonstrations are company-selected, and no public model, technical paper or independent test is available. The natural-sounding voice matters because voice realism can make impersonation harder to recognize, but the demonstrations do not establish how well the model works or whether it can imitate a specific person.
Block announced Dessa, a speech-to-speech research model, on October 23, with recordings that show it laughing, pausing and interrupting during conversation. The company says the model was trained from scratch on human conversation alone, but no public release, technical paper or independent evaluation is available, leaving its performance and safeguards unverified.
Block, the company behind Square and Cash App, says Dessa can listen and speak at the same time, respond to emotional cues and produce laughter, sarcasm and pauses. Its project page says the model was trained without text or pretrained models and that inference runs faster than real time on a MacBook Pro M4 Max or better. Block has not provided a specific latency figure.
The clips on Block’s project page feature a range of exchanges, including a conversation about a forgotten film title and a response to an anniversary surprise. Block says each is a live recording and that nobody steered the model. These remain company-provided demonstrations, not a substitute for outside testing. The Verge described some exchanges as “a little awkward and uncanny,” while reporting that Dessa sounded more natural than the other model in one comparison.
Block has not announced publicly available model weights or an API. Its current access is a waitlist and early access for a small group of AI safety researchers, according to the supplied source material. Technical details including the model’s size, benchmark results, license and public release date have not been disclosed.
Infographic 01 · Dessa fact sheet
What Block claims, what was heard, what is missing.
Every capability below comes from Block or its selected recordings. Nobody outside Block has run the model yet.
| Topic | What we know | Status |
|---|---|---|
| Training data | Human conversation only: “no text, no pretrained models” | Block’s claim |
| Conversation | Full-duplex: listens while it talks, backchannels, cuts in | Block’s claim |
| Emotion | Hears and speaks laughter, sarcasm, sighs; unclear if steered by tags (TechCrunch) | Block’s claim |
| Speed | “Faster than real time”; no latency figure | Block’s claim |
| Hardware | Trained on one NVIDIA GB200 NVL72; runs on a MacBook Pro M4 Max or better | Block’s claim |
| Demos | Some chats “a little awkward and uncanny”, more natural than GPT-Live-1 (The Verge) | reporters heard it |
| Size, benchmarks, release date | Not disclosed (CoinDesk) | not disclosed |
| Paper, model card, license, weights | None found as of 26 Oct 2026 | not disclosed |
| Languages | Not listed; Block calls learning new languages an open question | not disclosed |
| Voice cloning safeguards | No statement on watermarking, consent checks or imitating a specific person | not disclosed |
Who can use it
research preview
- Public: a waitlist; “launch links are coming soon”
- Early access: “a small group of AI safety researchers”, institutional email required
- Business use: none announced; Square already uses voice AI for ordering (TechCrunch)
Who built it
from Block’s page
- A “(very) small AI research lab at Block” reporting to Jack Dorsey
- Founded by former members of Dessa, the Toronto startup Square acquired in 2020 (CoinDesk)
- Lead Ragavan Thurairatnam: “a GPT3 moment for speech”
Sources: Block, Dessa project page (23 Oct 2026); Ragavan Thurairatnam on X (23 Oct 2026); The Verge (23 Oct 2026); TechCrunch (24 Oct 2026); CoinDesk (24 Oct 2026).
Powered by Thorsten Meyer AI
Natural Speech Raises Trust Risks
Conversational timing can make a synthetic voice feel more familiar and responsive. That may improve voice interfaces, but it can also make it harder for a listener to spot an artificial caller from conversational cues alone. Realistic speech is not proof of identity, and Block’s demonstrations do not show that Dessa has been used in fraud.
The concern sits alongside documented voice-cloning scams. The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI and $893 million in reported losses. The source material says the FBI report also described more than $5 million in losses from distress scams involving cloned voices of loved ones. These figures cover reported complaints and losses; they do not establish that Dessa contributed to them.
For people and organizations, the practical point is broader than one model: a familiar-sounding voice should not be treated as sufficient verification for urgent requests or money transfers. A callback to a known number and a pre-agreed family verification phrase can provide a separate check.
Products Worth Considering
Sony WH-1000XM6 Wireless Noise-Canceling Over-Ear Headphones, Blue, Bundle
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Bose QuietComfort Headphones (2nd Gen), Wireless Noise-Cancelling Over-Ear Headphones, Eucalyptus Green
As an affiliate, we earn on qualifying purchases.
What Block Has Disclosed
Block introduced Dessa through a project page on October 23, 2026. Co-founder Jack Dorsey described talking with the system as “like talking with a person,” citing emotion, nuance, pauses and cadence. Dessa lead Ragavan Thurairatnam called it “a GPT3 moment for speech,” according to the source material. Those are company and team characterizations, not independent findings.
Block says the model learned from hundreds of times fewer words than large language models and was trained using one NVIDIA GB200 NVL72. It has not published enough technical information to independently assess those claims. The company’s project page also frames learning new languages efficiently as an open question.
Outside coverage has emphasized the evidence gap. TechCrunch reported that it was unclear whether emotion in the demonstrations was guided by tags. CoinDesk noted the absence of model size, benchmark results and a release date. A separate question remains whether a system that selects its own voice and personality can be made to imitate a specific person; the source material says Block has not provided a clear account of that capability.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Jack Dorsey, Block co-founder
Testing and Safeguards Remain Open
As of October 26, no outside researcher or reporter is reported to have run the model directly. Without a paper, model card, benchmarks or public access, its conversational reliability, language coverage and hardware requirements cannot be checked independently. Block has not published a release date or license.
The available material also does not settle whether emotional delivery is generated directly or steered by tags, or what controls exist for consent, watermarking and person-specific voice imitation. Block says Dessa picks its own voice and personality, but that does not answer whether users can make it reproduce another person’s voice. The demonstrations show selected interactions, not how the model behaves across ordinary use or under safety testing.
Watch for Independent Access
The next substantive evidence would be independent access to Dessa, a technical paper or model card, and published results that explain how performance was measured. Those disclosures could clarify the model’s capabilities, limitations, supported languages and safeguards. Block has not announced when any of them will be available.
Until then, the company’s recordings support a limited conclusion: Block has presented a research system that appears conversational in selected clips. They do not establish general reliability, emotional understanding or protection against voice impersonation. People handling urgent calls can use established verification practices now, without waiting for this model to reach wider access.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says can listen and respond in real time, using conversational features such as pauses, laughter and interruptions.
Can the public use Dessa?
Not according to the information available as of October 26, 2026. Block has described a waitlist and limited early access for AI safety researchers, but has not announced public model weights or an API.
Has an independent test confirmed Block’s claims?
No independent test is reported in the source material. Block’s demonstrations are company-selected recordings, and no public technical paper, model card or benchmark results have been released.
Does Dessa show that AI can imitate a particular person?
The demonstrations do not establish that. Block says the system chooses its own voice and personality, but the available information does not clearly explain whether it can imitate a specific person or what safeguards govern that use.
How can people verify a suspicious voice call?
Do not rely on a familiar voice alone. For an urgent request, hang up and call back using a number you already know; families can also agree on a private verification phrase.
Source: ThorstenMeyerAI.com
