Does A Voice That Hesitates Sound More Human? Block’s Dessa Offers A Test

🔍 Read the full analysis: Does A Voice That Hesitates Sound More Human? Block’s Dessa Offers A Test on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get smart everyday buys delivered free with Prime

• Fast, free delivery on millions of items
• Prime Video, Amazon Music and more included
• Member-only deals all year

Start your free Prime trial →Free trial for eligible customers · Cancel anytime

As an affiliate, we earn on qualifying purchases.Hide

Block announced Dessa, a research model that listens and responds in spoken conversation, on October 23, releasing company-selected recordings that feature laughter, interruptions and emotional-sounding speech. The demonstrations have drawn attention both for their conversational realism and for the risk that realistic voice AI could make impersonation scams harder to recognize, but the model has not been independently tested and is not publicly available.

Block says Dessa was trained from scratch on recorded human conversation, 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 use conversational cues such as laughter or sighs. Its project page includes recordings of the system laughing mid-sentence, helping a speaker recall a film title and reacting to a surprise anniversary gift.

Block says the model runs faster than real time on a MacBook Pro M4 Max or better and was trained using one NVIDIA GB200 NVL72. The company has not supplied a latency figure, model size, benchmark results or a technical paper. Co-founder Jack Dorsey described talking with Dessa as “like talking with a person,” while project lead Ragavan Thurairatnam called it “a GPT3 moment for speech.” Those are company-side assessments, not results from independent evaluation.

Access remains limited. Block has offered a waitlist and early access to a small group of AI safety researchers, but has announced no public release date, weights or API. The Verge, after listening to the demonstrations, described some exchanges as “a little awkward and uncanny,” while also finding Dessa more natural than another model used in a comparison. TechCrunch reported that it was unclear whether some emotional qualities were guided by tags.

More Natural Voices, Greater Fraud Risk

Dessa’s potential relevance extends beyond the quality of a conversation. Voice impersonation scams already exist; a system that handles interruptions and sounds emotionally responsive could make it harder for a listener to rely on conversational awkwardness as a warning. The source cites FBI figures from the 2025 Internet Crime Report, published in April 2026: the bureau recorded 22,364 complaints mentioning AI and $893 million in reported losses. It also counted more than $5 million in losses from distress scams involving cloned voices of loved ones. These figures describe reported complaints and losses, not cases attributed to Dessa.

The same qualities could also make voice interfaces more accommodating. Block says its model can respond to emotion rather than treating speech as words alone. But the practical benefits and fraud risks are both still prospective: there is no public model to test, and the demonstrations do not establish how the system performs across users, situations or attempts to imitate a particular person.

For readers, the immediate point is not that Dessa has caused fraud; no such link is reported. It is that familiar voice cues are already a weak basis for verifying identity, and increasingly fluid AI conversations could make that weakness more consequential. For any request involving money or urgent action, independent verification matters more than whether the voice sounds familiar.

From Block Research to Public Demos

Block, the company behind Square and Cash App, published its Dessa project page on October 23, 2026. The project team is described as a small research lab at Block founded by former members of Toronto-based deep-learning startup Dessa, which Square acquired in 2020, according to CoinDesk. The announcement presents the model as a research effort, not a product launch with general customer access.

The release relies on audio selected and published by Block. The company says each clip is a live recording of a person talking with Dessa and that nobody steered the conversation. That description has not been independently verified. Reporters have listened to the clips, but no outside researcher or journalist has publicly tested the model itself, because access remains restricted.

Block frames the work as an alternative to speech systems that convert audio into text and then generate a response. It argues that such pipelines can discard emotional information. Reviewers have focused instead on the limits of what the clips establish: The Decoder cautioned that sounding emotional does not establish emotion or intelligence. The evidence available so far is therefore about how selected recordings sound, not a validated account of broad capability.

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

— Jack Dorsey, Block co-founder

What the Recordings Cannot Establish

Dessa’s capabilities remain unverified outside Block’s demonstrations. The company has not published a research paper, model card, benchmarks, weights or license, and reporters have not been given general access. It is not clear how reliably the system handles ordinary conversations, noisy calls, different accents or languages, or how its performance compares under standardized tests.

Block has not disclosed whether emotional expression is produced entirely by the model or partly guided by prompting or tags. The source material also reports no clear statement on watermarking, consent checks or protections against imitating a specific person. Block says Dessa chooses its own voice and personality, but that does not answer whether users could direct it to reproduce another person’s voice. The company’s claim that the recordings were not steered has not been independently confirmed.

It is also unknown when, or whether, the model will be released beyond the limited researcher access. The extent of language support is an open question, and no public timetable or product plan has been announced. The clips show examples of interaction; they do not establish that Dessa is generally human-like, safe from misuse or responsible for any reported fraud.

Release Details and Safety Testing

The next meaningful developments would be independent researcher access, technical documentation and repeatable evaluations of the model’s performance and safeguards. Block has said a small group of AI safety researchers has early access, but it has not announced when their findings might become public or whether broader access will follow.

Until those details emerge, companies and households should not treat a familiar or emotionally convincing voice as proof of identity. For financial or urgent requests, use a callback to a number already on file and verify through a separate channel. Families can agree on a private code word and hang up to call a known number if a supposed relative requests immediate help. Those precautions address voice impersonation generally; they are not evidence that Dessa itself is being used in scams.

Key Questions

What is Block’s Dessa?

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

Can the public use or test Dessa?

Not generally, as of October 26, 2026. Block has described a waitlist and limited early access for AI safety researchers, but has announced no public release date, API or model weights.

Has an independent test confirmed Block’s claims?

No public independent test of the model is reported. Journalists have reviewed Block’s selected recordings, but the model’s benchmarks, technical documentation and performance on broader tests have not been published.

Does Dessa prove that voice AI is being used in these scams?

No. The source reports existing voice-cloning distress scams and FBI loss figures, but it does not link those cases to Dessa. The concern is that more natural voice systems could make impersonation harder to detect.

How can people check an urgent call from a loved one?

Do not rely on the voice alone. Hang up and call a number you already know, verify the request through another channel, and consider using a family code word agreed on in advance.

Source: ThorstenMeyerAI.com