🔍 Read the full analysis: Block’s Dessa Makes AI Speech Feel Human. The Voice Still Proves Nothing on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Block announced Dessa on October 23, describing it as a speech-to-speech research model trained on human conversation and able to listen while speaking. The demos suggest natural-sounding exchanges, but outside researchers have not tested the model, and Block has not published a paper, benchmarks or public access. Its humanlike conversational cues also raise questions about voice impersonation and authentication.
Block introduced Dessa, a speech-to-speech research model that the company says can listen and talk at the same time, with conversational cues such as laughter, pauses and interruptions. The October 23 announcement matters because natural-sounding voice interaction could improve voice assistants, but it may also make voice impersonation harder to identify; no outsider has publicly tested the model.
Block says Dessa was trained from scratch using human conversation only, without text or pretrained models. The company describes it as full-duplex: it can listen while speaking, cut in and respond with backchannel signals. Its project page says inference runs faster than real time on a MacBook Pro M4 Max or better, but gives no latency figure or detailed technical specifications.
The available evidence is limited to recordings selected and published by Block. The clips include laughter mid-sentence, a conversation about a forgotten film title and a response to an anniversary surprise. Block says the recordings are live conversations without anyone steering the model. The company has not released model weights or an API, and access is described as a waitlist and early access for a small group of AI safety researchers.
Independent reporting has described both natural exchanges and awkward moments. The Verge reported that some conversations were “a little awkward and uncanny,” while finding Dessa more natural-sounding than a model used for comparison. TechCrunch said it was unclear whether emotional expression was being guided by tags. As of October 26, Block had published no technical paper, model card or benchmark results, leaving the model’s capabilities difficult to assess independently.
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
Humanlike Voices and Fraud Checks
The same conversational qualities that may make an assistant easier to talk to—laughter, emotional tone and natural turn-taking—could make a synthetic voice less obvious to a listener. That is a risk to monitor, not proof that Dessa has been used for fraud. Voice cloning and impersonation scams already exist without this model.
The source material cites the FBI’s 2025 Internet Crime Report, published in April 2026, which recorded 22,364 complaints mentioning AI and $893 million in reported losses. The same report attributed more than $5 million in losses to distress scams using cloned voices to mimic loved ones. Those figures cover reported complaints and losses; they do not establish that Dessa contributed to any of them.
For organizations, the practical issue is whether a familiar-sounding voice can safely authorize a payment or account change. A voice should not stand alone as proof of identity. For families, a call that sounds emotional or urgent is not enough to verify who is speaking. Callbacks to a known number and an agreed family code word can provide an independent check.
Popular right now
TCL 85 Designer Series QD-Mini LED TV (Model 85A400U) with Google Smart TV
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
Dyson V8 Cordless Vacuum
As an affiliate, we earn on qualifying purchases.
What Block Has Demonstrated So Far
Block, the company behind Square and Cash App, announced Dessa through a project page on October 23, 2026. Co-founder Jack Dorsey promoted the model’s conversational style, while Dessa lead Ragavan Thurairatnam described its potential in ambitious terms. These are statements from people associated with the project, not independent evaluations.
The project is being developed by a small research team at Block. The source material says the team includes former members of Dessa, a Toronto deep-learning startup acquired by Square in 2020. Block says the model was trained using one NVIDIA GB200 NVL72 and learned from hundreds of times fewer words than large language models require. Without a paper, training details or benchmarks, those comparisons remain company claims that cannot yet be checked.
Block has framed the work as a way to make voice AI more responsive to emotion, arguing that conventional speech-to-text systems can lose signals such as pain or frustration. Critics cited in the source material have raised the countervailing concern: a voice that sounds convincingly human could lower a listener’s guard. Both arguments concern the same feature—more natural expression—but neither establishes how Dessa will perform in routine use.
“Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.”
— Jack Dorsey, Block co-founder
Independent Tests and Safeguards Missing
It remains unclear how well Dessa performs beyond Block’s selected clips. The company has not published a paper, model card, benchmark results, model size or license, and no public release date has been announced. The available reports do not establish how the model behaves across longer conversations, different speakers or ordinary noisy conditions.
Block has not provided a list of supported languages. The project page itself identifies learning new languages efficiently as an open question. The source material also found no public explanation of watermarking, consent checks or whether the model can imitate a specific person. Block says Dessa chooses its own voice and personality, but that statement does not resolve how voice identity is protected or what restrictions may apply.
There is also no independent confirmation of the company’s claims about training data, compute requirements or real-time performance. The demonstrations can show what happened in those recorded interactions, but cannot by themselves establish how consistently the model behaves or how it could be misused.
Awaiting Access and Technical Details
The next useful evidence would be independent access, technical documentation and repeatable tests. Block has described early access for a small group of AI safety researchers, but has not said when broader access, a public release or model documentation will arrive. Until those details are published, claims about performance and safeguards remain difficult to verify.
In the meantime, businesses can review identity checks for payment and account requests, using a known callback number or a separate approval channel rather than relying on a caller’s voice alone. Families can agree on a code word and verify urgent requests through a number they already trust. Those precautions address voice impersonation generally; they are not evidence that Dessa has been used in a scam.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says can listen and speak at the same time. The company says it was trained from scratch on human conversation rather than text or pretrained models.
Can the public use Dessa?
Not based on the information available as of October 26, 2026. Block has described a waitlist and early access for a small group of AI safety researchers, but has announced no public API, model weights or release date.
Has anyone independently verified the model’s abilities?
No public independent test is described in the source material. The evidence consists of Block’s demonstrations and media reports that listened to them; Block has not published benchmarks or a technical paper.
Does Dessa prove that a human-sounding voice is genuine?
No. A natural-sounding voice is not proof of identity. Voice impersonation scams predate this announcement, so people and businesses should verify urgent requests through a known callback number or another independent method.
Source: ThorstenMeyerAI.com
