🔍 Read the full analysis: Block’s Dessa, Human-Sounding AI, And The Limits Of What We Can Hear 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, a speech-to-speech research model that it says was trained only on recorded human conversation and can listen and respond at the same time. Its demonstrations show expressive, conversational speech, but the model has not been independently tested and Block has not released a paper, benchmarks, weights or an API. The claims arrive as authorities report substantial losses from AI-related fraud, sharpening concerns about voice impersonation.
Block announced Dessa, a speech-to-speech research model, on October 23, with recordings in which it laughs, responds to interruptions and speaks with apparent emotion. The company says it was trained from scratch on human conversation alone, but no independent party has tested the model, and Block has not published a technical paper, benchmarks or a public release date—leaving both its capabilities and its risks difficult to assess.
Block, the company behind Square and Cash App, says Dessa can listen while speaking, interrupt or provide conversational responses, and recognize vocal cues such as laughter, sarcasm and frustration. The company describes it as a model trained without text or pretrained models. It says inference runs faster than real time on a MacBook Pro M4 Max or better, but has not provided a latency figure or independently verifiable test results.
The demonstrations on Block’s project page include a model laughing mid-sentence, helping a speaker recall a film title and responding to an anniversary surprise. Block says the recordings are live conversations and that nobody steered the model. Those recordings are company-selected examples, however, not an independent evaluation. The Verge reported that some exchanges felt awkward or uncanny, while TechCrunch noted that it was unclear whether emotion in the output was controlled by tags.
Block has not released model weights or an API. As of October 26, access is through a waitlist, with early access for a small group of AI safety researchers, according to the source material. The company has not provided a model card, license, benchmark results, full details of the model’s size or a release schedule. Its project page also leaves open how well the system learns other languages.
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-Like Speech and Voice Fraud
Dessa’s demonstrations matter because the conversational cues that can make voice assistants easier to talk to can also make an impersonation harder to recognize. Natural timing, laughter and interruptions are not proof of human identity; they can become part of synthetic speech. That does not establish that Dessa can clone a particular person or has been used in fraud. Block has not publicly answered whether it can imitate specific voices or what safeguards it includes.
The risk exists beyond this model. The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI and $893 million in reported losses. It also reported more than $5 million in losses from distress scams using cloned voices of loved ones, and more than $30 million in losses from AI-assisted business email compromise scams. These figures describe reported complaints and losses; they do not establish that Dessa was involved.
For organizations, the development is a reminder that a familiar-sounding caller should not be treated as proof of identity, particularly when money or account access is involved. Verification through a known callback number and written approval processes can provide checks that voice recognition alone cannot. For families, agreeing on a code word and calling back using a saved number are practical precautions against urgent impersonation calls.
Products Worth Considering

Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black, International Version

Apple AirPods Pro 3 Wireless Earbuds with Active Noise Cancellation

Bose Smart Ultra Dolby Atmos Soundbar with Voice Control, HDMI eARC, Black
As an affiliate, we earn on qualifying purchases.
Block’s Speech-Only Research Claims
Block describes Dessa as the work of a small research team associated with Dessa, the Toronto deep-learning startup acquired by Square in 2020. The project page says the model learned from human conversation rather than text and was trained using one NVIDIA GB200 NVL72. Block has not supplied enough technical detail for outside researchers to verify the training account or compare its resource claims with other systems.
Co-founder Jack Dorsey promoted the conversational quality of the system, while project lead Ragavan Thurairatnam described it as “a GPT3 moment for speech” and said it could run on consumer GPUs. Those are company-linked characterizations, not results from a published benchmark. The audio clips provide a way to hear the system, but they do not establish how it performs across users, languages, noisy environments or extended conversations.
The launch has drawn both enthusiasm about more responsive voice interfaces and concern about deception. Block’s stated case is that voice systems should respond to emotion rather than ignore it. Critics have questioned why generated voices need to sound human. Those reactions point to a central tension: the same realism can improve interaction and make synthetic speech more persuasive.
“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
The main uncertainty is how Dessa performs outside Block’s demonstrations. No independent researcher or reporter has publicly tested the model directly, and the company has not released a technical paper, model card, benchmark results, weights or API. The selected clips cannot establish consistency, error rates, latency in varied conditions or whether the model’s conversational behavior holds up in longer interactions.
Other unanswered questions include which languages it supports, whether vocal emotion is generated naturally or guided by tags, and what measures exist to prevent misuse. Block says the system chooses its own voice and personality, but that does not resolve whether it can imitate a specific person. No evidence in the source material shows Dessa has been used in a scam, and its ability to clone an individual’s voice is not established.
Watch for Research and Release Details
The next meaningful development would be technical documentation or access for independent evaluation. That could clarify how the system was trained, what it can do across languages and whether safeguards or detection measures are built in. Block has not announced when broader access, model weights or an API might become available, so no public release date can be reported.
Until more evidence is available, readers should treat the demos as demonstrations of Block’s claims, not proof of general performance. Companies handling payments can use independent identity checks for unusual requests, while individuals can end an unexpected urgent call and reconnect through a number they already trust. The key test ahead is whether outside evaluation supports the launch claims and makes the model’s limitations and protections clear.
Key Questions
What is Block’s Dessa?
Dessa is a speech-to-speech research model that Block says was trained from scratch on human conversation. The company says it can listen and speak at the same time and respond to vocal cues.
Has Dessa been independently tested?
No public independent test is reported in the source material. Block’s available evidence consists of recordings selected by the company, and it has not published benchmarks or a technical paper.
Can Dessa imitate a specific person?
That has not been established. The source material says Block has not clarified whether the model can imitate a specific person or what safeguards it uses to prevent misuse.
Is Dessa publicly available?
No public release has been announced. As of October 26, 2026, access is through a waitlist, with early access for a small group of AI safety researchers, according to the source material.
How can people respond to suspicious voice calls?
Do not treat a familiar voice as sufficient proof of identity. Hang up and call back using a number you already know; families may also agree on a code word for urgent requests.
Source: ThorstenMeyerAI.com
