A Person-Like Voice, An Open Question: Block’s Dessa And AI

🔍 Read the full analysis: A Person-Like Voice, An Open Question: Block’s Dessa And AI 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 speech-to-speech research model that the company says can listen while speaking, respond to interruptions and produce cues such as laughter and sighs. The release matters because realistic conversational voices could make voice-based services feel more natural, while also making it harder for people to recognize an impersonation; no outside researcher has publicly tested the model.

Block published a project page for Dessa on October 23. The company says it trained the model from scratch on human conversation audio alone, without text or pretrained models. It describes the system as “full-duplex”: it can listen while it talks, rather than waiting for a complete spoken request before responding. Block says that lets it decide when to interrupt or provide conversational responses such as brief acknowledgments.

The company’s recordings feature laughter during speech, a listener helping with a forgotten film title, and a reaction to an anniversary surprise. Block says the clips are live conversations and that no one steers the system during them. These demonstrations are company-selected examples, not results from an independent evaluation. The Verge described some exchanges as “a little awkward and uncanny,” while finding Dessa more natural-sounding than another model used for comparison.

Block says Dessa runs faster than real time on a MacBook Pro M4 Max or better and was trained using one NVIDIA GB200 NVL72. It has not published a specific latency figure, model size, benchmark results or technical paper. Access is limited to a waitlist and a small group of AI safety researchers, according to the supplied account; Block has not announced public model weights or an API.

Humanlike Voices and Fraud Risks

Dessa’s significance lies not in proof that it can deceive people, but in the kind of interaction Block is trying to build. A voice that responds to pauses, overlaps and emotion may make a legitimate automated conversation less rigid. The same qualities could also remove some cues people use to suspect a synthetic caller. That risk exists beyond Dessa: voice cloning and impersonation scams have already been reported.

The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI and $893 million in reported losses across those complaints. It separately reported more than $5 million in losses to distress scams involving cloned voices of loved ones. Those figures describe complaints and reported losses, not the performance of Dessa or a forecast of future harm.

For organizations, the development is a reminder that recognizing someone by voice alone is not a reliable safeguard for high-stakes requests. A familiar-sounding caller asking for a payment or account change should be verified through a separate channel. For families, an agreed code word and calling back on a known number can help check urgent claims. These steps address existing impersonation risks; they are not evidence that Dessa itself has been used in fraud.

How Block Presented Dessa

Block is the company behind Square and Cash App. Its small research team says it built Dessa as a speech-first system rather than a text model connected to a speech generator. Block argues that conventional speech-to-text systems can discard information such as tone and laughter, and says Dessa is designed to hear and produce those signals directly.

Block co-founder Jack Dorsey wrote that talking with the model “feels like talking with a person,” citing emotion, pauses and natural cadence. Dessa lead Ragavan Thurairatnam called it “a GPT-3 moment for speech,” and said it could run on consumer GPUs. These are the team’s characterizations, not independently established assessments of capability.

Other coverage has pointed to unanswered technical questions. TechCrunch said it was unclear whether emotional delivery might be steered by tags. CoinDesk reported that the company had not disclosed the model’s size, benchmark results or a public release date. The Decoder cautioned that sounding emotional does not establish that a system is emotional or intelligent. The available evidence remains Block’s descriptions and recordings, plus reporters’ impressions of those clips.

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

— Jack Dorsey, Block co-founder

The Evidence Behind the Demos

Dessa has not been independently evaluated in the material available as of October 26. There is no public technical paper, model card, benchmark suite, license or release date, and outsiders have not run the model. The recordings cannot establish how consistently it handles interruptions, difficult conversations or unfamiliar voices outside the scenarios Block chose to publish.

Other unresolved issues include which languages it supports, how its emotional delivery is produced, and whether safeguards can prevent a user from making it imitate a specific person. Block says the model selects its own voice and personality, but that does not answer every question about voice imitation, consent, disclosure or watermarking. The company has not provided those details in the supplied material. Nor is there evidence that Dessa has been used in a scam.

Access, Testing and Safeguards

The next meaningful developments would be independent access and technical documentation that allow researchers to assess performance, limitations and safeguards. Block has described a waitlist and early access for a small group of AI safety researchers, but has not set a date for wider availability. Until testing is possible, claims about the system’s reliability, language range and resistance to misuse should be treated as unverified company statements.

Meanwhile, people and organizations can respond to the broader risk without waiting for Dessa to launch publicly: verify unusual financial requests by calling a number already on file, require a second approval for significant transfers, and agree on a family check phrase for emergencies. The source account also points to the EU AI Act’s synthetic-content transparency rules, which it says have applied since August 2, 2026. The precise obligations depend on the use case; Block has not announced Dessa-specific disclosure or consent controls.

Key Questions

What is Block’s Dessa?

Dessa is a speech-to-speech research model that Block says was trained on human conversation audio and can listen while speaking. It is not publicly available for general use as of October 26, 2026.

Has an independent researcher tested Dessa?

Not publicly, based on the information available. The demonstrations are recordings published by Block; there is no public technical paper or benchmark report in the supplied account.

Does Dessa clone a specific person’s voice?

That remains unclear. Block says Dessa picks its own voice and personality, but has not detailed whether it can imitate a named person or what consent checks may apply.

Does the announcement show that Dessa has been used for fraud?

No. The source material reports no evidence of Dessa being used in a scam. It discusses broader concerns because voice impersonation scams already exist and humanlike speech could make synthetic callers harder to identify.

How can people check a suspicious voice call?

Hang up and call the person or organization back using a number already known to you. For family emergencies, a pre-agreed code word can help verify a caller’s identity; urgency should prompt extra checking, not an immediate transfer.

Source: ThorstenMeyerAI.com