🔍 Read the full analysis: How Block’s Dessa Uses Human-Sounding Speech Without Proving It’s Human on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Block announced Dessa, a speech-to-speech research model that it says was trained on recorded human conversation and can listen and speak at the same time. The company’s selected demos sound expressive, but outside researchers and reporters have not tested the model, and technical details and public access remain limited.
Block has published demonstrations of Dessa, a research model that it says can listen and speak at the same time, interrupt naturally and reproduce features such as laughter and emotional inflection. The October 23 announcement puts a spotlight on a safety question for voice technology: the company’s recordings suggest a more conversational style, but no independent testing or public model access is available to establish how well it works or how it could be misused.
Block describes Dessa as a speech-to-speech model trained from scratch on human conversation, without text or pretrained models. The company says it handles speech directly rather than converting a speaker’s words into text before generating a reply. That design, Block argues, preserves cues such as emotion, sarcasm, laughter and pauses that a text-based pipeline may lose.
In demonstrations hosted by Block, the system laughs during a conversation, responds to a story about a surprise anniversary and helps a speaker recall a film title. Block says the clips are live recordings and that nobody steered the model during them. Those are company-provided examples, not independent evaluations. The Verge reported that the conversations could sound awkward or uncanny, while also describing Dessa as more natural than another model it compared it with.
Block says inference runs faster than real time on a MacBook Pro M4 Max or better, and that the model was trained using one NVIDIA GB200 NVL72. The company has not published a latency figure, model size, benchmark results, technical paper or model card. It has not announced a public API or release of model weights; access is limited to a waitlist and a small group of AI safety researchers, according to the source report.
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
Human-Sounding Voices and Fraud Risk
A system that handles interruptions, pauses and emotional cues could make automated conversations feel less mechanical. That may be useful in customer service or other voice interfaces, but the same qualities could make it harder for a listener to distinguish a synthetic voice from a person. Natural delivery is not proof of identity, and Block’s demonstrations do not establish that Dessa can imitate a particular person.
The concern sits alongside existing voice-cloning fraud, rather than depending on Dessa itself. The source report 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 FBI report also recorded 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 show that Dessa was involved.
For businesses, the development is a reminder that a familiar-sounding voice should not be treated as a standalone authorization method for payments or sensitive account changes. Callbacks to known numbers and separate verification steps can reduce reliance on voice recognition. For families, a private code word and independently calling the person back are practical precautions against urgent requests delivered over the phone.
Products Worth Considering
Logitech MX Master 4 for Mac, Ergonomic Bluetooth Mouse with Haptic Feedback, Ultra-Fast Scrolling, USB-C Charging, for MacOS, iPadOS, Windo
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
Sony WH-1000XM4C Wireless Noise-Cancelling Headphones, Black
As an affiliate, we earn on qualifying purchases.
What Block Has and Has Not Released
Block, the company behind Square and Cash App, posted its Dessa project page on October 23, 2026. Co-founder Jack Dorsey described speaking with the model as feeling like talking with a person, citing emotion, nuance, pauses and cadence. Dessa lead Ragavan Thurairatnam called the work “a GPT3 moment for speech” and said it could run on consumer GPUs. These are the company team’s characterizations, not conclusions from a published evaluation.
The model’s name refers to the Toronto deep-learning startup Dessa, whose former members founded the research team at Block; Square acquired that startup in 2020, according to CoinDesk. Block’s stated aim includes preserving emotional information in speech. The company says present voice AI can miss cues such as a caller being in serious pain. That potential benefit and the risk of making impersonation more convincing arise from the same ability to respond to how a voice sounds.
Coverage by The Verge, TechCrunch and CoinDesk has pointed to gaps in the public evidence. TechCrunch said it was unclear whether emotional delivery might be guided by tags, while CoinDesk noted the lack of disclosed model size, benchmark results and release date. The available demonstrations cannot answer those questions, or show how the model performs across speakers, languages or difficult real-world calls.
“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 Are Still Missing
As of October 26, there is no public technical paper, model card, benchmark suite or downloadable model identified in the source material. No outside reviewer has run Dessa directly. It is not yet clear how the system performs outside the published conversations, how reliably it handles interruptions, or whether the emotional qualities in the clips are generated without additional steering.
Block has not listed supported languages or given a public release date. The company says the model picks its own voice and personality, but the available information does not settle whether it can reproduce a specific person’s voice, what protections against that use may exist, or whether outputs will be watermarked. Those gaps make it too early to assess the model’s capabilities and safeguards independently.
Awaiting Access and Technical Detail
Block has said that a small group of AI safety researchers will receive early access, but the source material gives no timetable for wider testing or a general release. The next useful evidence would include a technical report, documented benchmarks, access for independent researchers and clear information about safeguards. Until those are available, claims about latency, language learning and conversational performance remain Block’s claims.
In the meantime, individuals and organizations can avoid treating a convincing voice as sufficient verification. For a money request or other high-impact instruction, use a callback to a number already on file and confirm through a separate channel. The reported demonstrations show what Block says its model can do; they do not establish that the voice is human or that any particular call is genuine.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says was trained on human conversation and can listen and talk at the same time. The company has shown recordings but has not made the model publicly available.
Has anyone outside Block tested the model?
Not directly, according to the source report. Media outlets have listened to Block’s demonstrations, but no outside party has run the model or published independent benchmark results.
Does Dessa prove that an emotional-sounding voice is human?
No. A voice that sounds natural or expressive is not proof of who—or what—is speaking. The demonstrations are company-provided recordings and do not verify a speaker’s identity.
Can Dessa clone a specific person’s voice?
That remains unclear. Block says Dessa chooses its own voice and personality, but the public information cited in the source does not explain whether it can imitate a specific person or what safeguards would prevent that use.
What should people do about suspicious voice calls?
Do not rely on a familiar voice alone for urgent requests. Hang up and call back using a number you already know, verify important instructions through another channel, and consider using a private family code word.
Source: ThorstenMeyerAI.com
