🔍 Read the full analysis: Block’s Dessa Brings Laughter And Hesitation To AI Voice. Is That Enough? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 that it says was trained from scratch on recorded human conversation and can respond with laughter, pauses and interruptions. The demonstrations are Block-selected recordings; the model has not been made publicly available or independently tested, and its natural-sounding voice raises questions about impersonation risks.
Block published demonstrations on October 23 of Dessa, an experimental speech-to-speech AI model that the company says can laugh, express emotion and interrupt naturally in conversation. The demos suggest a more humanlike style of voice interaction, but the model is not publicly available and has not been independently tested, leaving both its capabilities and its implications for voice fraud uncertain.
Block says Dessa was trained from scratch using human conversation audio only, without text or pretrained models. The company describes it as a full-duplex system: it can listen while speaking, respond to pauses and interruptions, and reproduce vocal cues such as laughter, sarcasm and a frustrated sigh. Block says it runs faster than real time on a MacBook Pro M4 Max or better, but has not published a latency figure or detailed technical documentation.
The recordings on Block’s project page include a model laughing mid-sentence, helping a speaker remember a film title and responding to a surprise-anniversary story. Co-founder Jack Dorsey wrote that talking with Dessa feels like talking with a person, citing its emotion, nuance, pauses and cadence. Block says the clips are live conversations and were not steered. They remain company-selected demonstrations, rather than results from independent testing.
As of October 26, Block had published no paper, model card, benchmark results, license or release date. The company has not announced public model weights or an API. Access is limited to a waitlist and early access for a small group of AI safety researchers. The model’s size, language coverage and safeguards against impersonation are not publicly detailed.
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 Voices Raise Fraud Stakes
A conversational voice that handles interruptions and sounds emotionally responsive could make legitimate voice interfaces feel easier to use. The same qualities can also make it harder for a listener to distinguish a generated voice from a real person, particularly during an urgent call. Voice alone is not reliable proof of identity, even without Dessa: voice cloning is already used in reported scams.
The FBI’s 2025 Internet Crime Report, published in April 2026, counted 22,364 complaints mentioning AI, with $893 million in reported losses. It also recorded more than $5 million in losses to distress scams using cloned voices of loved ones. Those figures cover reported complaints and losses, not every incident, and do not establish that Dessa was used in any fraud. They show why advances in natural-sounding speech are relevant to consumers and businesses.
For organizations, a convincing voice should not substitute for a second verification step on payment or account changes. For families, independently calling someone back using a number already known to them can help check an urgent request. These are practical precautions against voice impersonation generally; the demonstrations do not establish that Dessa can clone a particular person.
Products Worth Considering
Apple MacBook Pro 14.2-inch with M3 Max, 14-Core CPU, 30-Core GPU, 36GB RAM, 1TB SSD, Space Black
Apple MacBook Pro 2023 14 with M3 Pro Chip, 18 GB RAM, 2 TB SSD, Space Black
Apple MacBook Pro 16.2-inch (M5 Pro, 48GB RAM, 2TB SSD, Space Black)
As an affiliate, we earn on qualifying purchases.
What Block Has Demonstrated
Block, the company behind Square and Cash App, announced the project on October 23. Its team describes itself as a small research lab at Block, founded by former members of Toronto-based AI startup Dessa, which Square acquired in 2020. The name refers to that lab’s earlier identity; the new model is presented as a research project, not a consumer product.
Block says the model learned from hundreds of times fewer words than large language models require and was trained using one NVIDIA GB200 NVL72. Those are company-provided descriptions; without a paper or benchmarks, outsiders cannot assess the training claims, compare performance or reproduce the results. Block also says the system selects its own voice and personality, but has not explained what that means technically.
Coverage of the launch has been more cautious than the company’s promotional description. The Verge reported that some exchanges sounded awkward or uncanny, while finding Dessa more natural than another model used in a comparison. TechCrunch said it was unclear whether emotional delivery might be guided by tags. The Decoder cautioned that sounding emotional does not establish that a system is emotional or intelligent.
“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 central unknown is how Dessa performs outside the recordings Block chose to publish. No outsider has publicly run the model, and there are no released benchmarks, technical paper or model card. It is also unclear how often its speech is awkward, how well it handles varied speakers and noisy conditions, or whether its claimed speed applies consistently across hardware and tasks.
Block has not provided a list of supported languages or a public explanation of consent checks, watermarking or restrictions on imitating specific people. The company says the model chooses its own voice and personality, but that does not answer whether users could direct it to sound like someone else. There is no public evidence that Dessa has been used in scams, and the reported FBI figures do not attribute losses to this model.
Release Details Will Shape the Debate
Block has not announced a public release date, API or model-weight availability. Its next substantive opportunity to clarify the project would be to publish technical documentation, evaluation results and details of access for the safety researchers receiving early access. Until then, claims about performance and safeguards remain difficult to verify.
Companies evaluating voice systems can review how they verify callers and approve payments, while consumers can avoid relying on a familiar-sounding voice alone when money or urgent requests are involved. The relevant test for Dessa is not just whether its demos sound natural, but how it performs under independent evaluation and what protections accompany any broader access.
Key Questions
What is Block’s Dessa?
Dessa is an experimental speech-to-speech AI model that Block says was trained from scratch on recorded human conversation. The company says it can listen and speak at the same time and respond with vocal cues such as laughter and pauses.
Can the public use Dessa now?
No public release, model weights or API had been announced as of October 26, 2026. Block said access was through a waitlist, with early access for a small group of AI safety researchers.
Has Dessa been independently tested?
Not publicly. Reports have assessed Block’s demonstrations, but no outsider has publicly run the model. There are no published benchmarks or technical paper available in the source material.
Does the model clone a specific person’s voice?
That is not established. Block says Dessa chooses its own voice and personality, but has not publicly explained whether it can imitate a specific person or what safeguards would prevent that.
What should someone do about an urgent call asking for money?
Do not treat a familiar-sounding voice as proof of identity. A cautious step is to hang up and call the person back using a number already saved or otherwise independently verified, then confirm the request through another channel.
Source: ThorstenMeyerAI.com
