Block’s Dessa Talks Like A Person, But Sound Isn’t Proof

🔍 Read the full analysis: Block’s Dessa Talks Like A Person, But Sound Isn’t Proof on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on tech for your team

• Business-only prices and quantity discounts
• Tax-exempt purchasing
• Multiple users, one account, clear invoices

Create a free Business account →Free to register

As an affiliate, we earn on qualifying purchases.Hide

Block announced Dessa, a speech-to-speech research model, on October 23, with demonstrations of a voice that laughs, responds to pauses and speaks over conversation. Those recordings show what the company says the system can do, but no outside researcher has tested the model, and Block has published no paper, benchmark results, model card or public access. The gap matters as lifelike speech can make phone interactions feel more natural while also making voice impersonation harder to spot.

Block says Dessa was trained from scratch on human conversation, without text or pretrained models. The company describes it as full-duplex: it can listen while speaking, respond to conversational cues and backchannel rather than waiting for a turn to finish. Block also says the model can produce or interpret emotion, including laughter and sarcasm, and that inference runs faster than real time on a MacBook Pro M4 Max or better. It has not provided a latency figure or enough technical detail to verify those claims independently.

The recordings on Block’s project page include laughter mid-sentence, an exchange in which a speaker forgets a film title, and a response to news of an anniversary surprise. Block says the clips are live recordings and that nobody steered the conversations. Those are company-selected demonstrations, not independent evaluations across users, tasks or conditions. The Verge described some exchanges as occasionally awkward and uncanny, while finding Dessa more natural-sounding than another model in one comparison. TechCrunch reported that it was unclear whether emotion in the output was guided by tags.

Block co-founder Jack Dorsey wrote that “Talking with it feels like talking with a person”, citing emotion, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called it “a GPT3 moment for speech” and said it could run on consumer GPUs. Those descriptions are attributed statements from the company’s launch, not independently established findings. Block says a small group of AI safety researchers has early access and is taking waitlist requests; it has not announced weights, an API or a public release date.

Why Natural Voice Raises Fraud Risks

Dessa’s demos matter because conversational realism can serve legitimate uses and lower the cues people rely on to identify a synthetic voice. A system that handles pauses, interruptions and emotion may make voice interfaces more comfortable, but those same qualities could make an impersonation call sound more plausible. That risk does not depend on Dessa being released: voice cloning and fraud attempts already exist.

The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI, with $893 million in reported losses. It also said victims lost more than $5 million to distress scams that used voice cloning to imitate a loved one. These figures describe complaints and reported losses in 2025; they do not establish that Dessa was involved or that every complaint concerned voice fraud. They do show why more natural speech attracts attention from fraud watchers.

For organizations, a familiar-sounding voice should not be treated as proof of identity when approving payments or sensitive account changes. Callback procedures using a known number and written confirmation can add verification beyond the call itself. Individuals can likewise agree on a family code word and independently call back when a request involves money or urgency. These are practical precautions, not evidence that Dessa itself is being used in scams.

From Block’s Demo to Public Access

Block, the company behind Square and Cash App, published the Dessa project page on October 23, 2026. Its team is described as a small research lab formed from people associated with Dessa, the Toronto deep-learning startup Block acquired in 2020. Block says the model was trained using one NVIDIA GB200 NVL72 and learned from far fewer words than large language models typically require; it has not disclosed the model’s size or supplied comparative benchmarks to substantiate the efficiency claims.

The launch follows a wider debate about whether synthetic voices should sound human. Block argues that emotion recognition could help voice systems respond more appropriately to people in pain or distress. Critics, including filmmaker Justine Bateman, have raised concerns that realism can make it easier to lower a listener’s guard. The competing arguments concern the same feature: more human-like interaction. The demos do not establish how reliably Dessa recognizes emotion, how it performs outside selected exchanges or whether it can imitate a particular individual.

Block’s project materials characterize Dessa as choosing its own voice and personality. As of October 26, the company had not published a technical paper, model card or license, nor identified supported languages. It has also not publicly detailed watermarking, consent checks or safeguards against voice imitation. Those omissions limit what outsiders can conclude from recordings alone.

““Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.””

— Block co-founder Jack Dorsey

What the Demos Cannot Establish

Independent performance remains unverified. No outside researcher has run Dessa, and Block has not released benchmarks, model weights or a public API. The available clips cannot show how the system performs across a broad set of speakers, languages, noisy calls or longer conversations. Block has not published a list of supported languages or a public release timetable.

It is also unclear whether emotion in the demos is generated directly by the model or influenced by tags or other controls, a point raised by TechCrunch. The company has not publicly explained its approach to watermarking, consent checks or attempts to imitate a specific person. Block’s statement that Dessa selects its own voice and personality does not, by itself, answer how the system would respond to requests to sound like someone else. No evidence in the source material links Dessa to actual fraud; the concern is about potential misuse of lifelike voice technology generally.

Awaiting Independent Research Access

The next significant step will be access that allows researchers to test the model beyond Block’s recordings. Block says a small group of AI safety researchers has early access, but it has not announced when broader access, a paper, model card or public release might follow. Independent findings could clarify performance, limitations and safeguards; until then, the launch claims should be treated as unverified company descriptions.

Companies and families do not need to wait for Dessa to review voice-based verification. Payment requests and urgent claims can be checked through a separate, trusted channel rather than by relying on how familiar a voice sounds. Block’s announcement has made the capabilities visible, but it has not established how widely the model will be available or what protections will accompany it.

Key Questions

What is Block’s Dessa?

Dessa is a speech-to-speech research model that Block says was trained from scratch on recorded human conversation. Block says it can listen while speaking and respond to cues such as pauses and emotion.

Has anyone independently tested Dessa?

Not publicly as of October 26, 2026. The available evidence consists of Block’s demonstrations and reporting by outlets that listened to those clips; no outside researcher has run the model.

Can Dessa be used to clone a specific person’s voice?

That has not been established. Block says the model selects its own voice and personality, but has not publicly detailed whether it can imitate a named person or what safeguards prevent such use.

Does the FBI data show that Dessa has been used in scams?

No. The FBI’s figures concern AI-related complaints and reported losses in 2025, before Dessa’s October 2026 announcement. The source material does not link the model to any scam.

How can people check an urgent call asking for money?

Do not rely on voice alone. Hang up and call back using a number already known to you, verify payment requests through another channel and consider agreeing on a family code word for emergencies.

Source: ThorstenMeyerAI.com