Block’s Dessa And The Question Behind Every Human-Sounding AI Voice

🔍 Read the full analysis: Block’s Dessa And The Question Behind Every Human-Sounding AI Voice on ThorstenMeyerAI.com

Before you orderOffer from Amazon

Get the latest gadgets 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, on October 23, showing recordings in which it laughs, responds to emotion and speaks over conversational pauses. The announcement matters because realistic, responsive voices may make impersonation calls harder to recognize, but the model has not been independently tested and is not publicly available.

Block says it trained Dessa from scratch on recorded human conversation, without text or pretrained models. The company describes it as full-duplex: it can listen while speaking, respond to interruptions and use conversational cues such as laughter or sighs. Block also says it runs faster than real time on a MacBook Pro M4 Max or better, but gave no latency figure.

The company’s project page contains recordings of exchanges including a forgotten film title and a response to an anniversary surprise. Block says the clips are live conversations and were not steered. Those are company-selected demonstrations, not results from an independent evaluation. The Verge reported that some exchanges sounded awkward or uncanny, while describing the model as more natural than another voice system it compared.

As of October 26, Block had published no technical paper, model card, benchmark results, weights or API. The company has offered a waitlist and said a small group of AI safety researchers would receive early access. It has not announced a general release date.

Voice Realism Meets Fraud Risk

Dessa’s demonstrations put a familiar trade-off in sharper focus: conversational timing and emotional cues can make voice interfaces easier to use, but may also make a synthetic speaker harder to distinguish from a person. That does not show that Dessa is being used for fraud. It does illustrate why natural-sounding speech is not proof of identity.

The FBI’s 2025 Internet Crime Report, published in April 2026, recorded 22,364 complaints mentioning AI and $893 million in reported losses. The FBI also counted more than $5 million in losses to distress scams involving cloned voices of loved ones. These figures cover reported complaints and losses; they do not measure Dessa’s impact or establish that the model contributed to those cases.

For organizations, the practical issue is verification: a familiar voice on a call should not by itself authorize a payment or disclose sensitive information. For families, agreeing on a private code word and calling back using a known number can add a check. These precautions address existing voice-cloning risks, not a demonstrated capability unique to Dessa.

What Block Has—and Has Not—Shown

Block, the company behind Square and Cash App, published Dessa’s project page on October 23. Co-founder Jack Dorsey described speaking with it as feeling like talking to a person. Dessa lead Ragavan Thurairatnam called the project “a GPT3 moment for speech,” according to the source material. Those comments are Block leaders’ assessments, not independent benchmark findings.

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 published enough technical detail for outsiders to verify the training description or compare performance. TechCrunch reported that it was unclear whether emotional qualities in the examples were influenced by tags; CoinDesk noted the absence of disclosed model size, benchmarks and a public release date.

Block describes Dessa as choosing its own voice and personality. That does not answer whether it can imitate a specific person, what safeguards govern its use or how consent is handled. The company’s claim that the model runs locally also does not establish what hardware, setup or restrictions would apply to other users.

“a little awkward and uncanny”

— The Verge, in its October 23 report

Independent Tests Are Still Missing

The available evidence consists mainly of recordings selected by Block. No outside researcher or reporter has run the model, and the company has not released a paper, model card or benchmark results. It remains unclear how it performs across ordinary conversations, noisy settings, different accents or languages, and how often its responses are awkward or inaccurate.

Block has not specified whether the system can imitate a particular individual, what consent checks or watermarking it uses, or what limitations will apply to early access. It is also unclear how the company’s local-computing claim translates into practical availability beyond the stated Mac hardware. The demonstrations do not establish that Dessa can clone voices or that it has been used in scams.

Release Details and Safety Checks

The next milestones are independent access and technical documentation. Block has said a small group of AI safety researchers will receive early access, but has not given a schedule for that access or for a broader release. Their ability to test the model, and whether findings will be published, remains unknown.

Readers should watch for Block’s disclosures on performance, language coverage, consent, impersonation safeguards and audio labeling. Until those details are available, the demos support a limited conclusion: Block has shown a conversational voice model it says can respond with human-like cues, but its capabilities and risks have not been independently established.

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. The company says it can listen while speaking and respond to emotional and conversational cues.

Can the public try Dessa?

Not as of October 26, 2026. Block has a waitlist and says a small group of AI safety researchers will receive early access, but it has not announced a public release date, weights or API.

Has an independent reviewer tested the model?

No outside testing is described in the available reporting. Reviewers have listened to Block’s recordings, but the model itself has not been independently evaluated.

Does Dessa clone a specific person’s voice?

That has not been established. Block says Dessa selects its own voice and personality, but has not explained whether it can imitate a particular person or what consent safeguards are in place.

How can people reduce the risk of voice impersonation scams?

Do not rely on a voice alone to confirm someone’s identity or authorize a payment. For an urgent request, hang up and call a number you already know; families can also agree on a private code word.

Source: ThorstenMeyerAI.com