🔍 Read the full analysis: How Block’s Dessa Blurs The Line Between Human-Sounding And Human on ThorstenMeyerAI.com
Get the little things that make your day 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 it says was trained only on recorded human conversation and can listen and speak at the same time. Its demonstrations have drawn attention for human-like timing and emotion, but no independent researcher has tested the model and Block has not published technical documentation or announced a public release.
Block has introduced Dessa, a speech-to-speech research model that the company says was trained from scratch on recorded human conversation and can listen and talk at the same time. The announcement matters because its demonstrations feature laughter, interruptions and other conversational cues that may make synthetic voices more convincing, but the model has not been independently tested and is not publicly available.
Block published a project page for Dessa on October 23. The company says the model uses no text or pretrained models in its training and can respond to emotion, sarcasm, laughter and changes in a speaker’s voice. It also says Dessa can recognize when to interrupt or provide brief conversational responses while another person is speaking.
The project page includes recordings in which the model laughs mid-sentence, helps a speaker recall a film title and responds to a story about an anniversary surprise. Block describes the clips as live recordings and says nobody steered the conversations. These are company-selected demonstrations, not results from independent testing.
Block says inference runs faster than real time on a MacBook Pro M4 Max or better, but it has not provided a specific latency figure. It also says the system was trained using one NVIDIA GB200 NVL72. Dessa is available only through a waitlist and early access for a small group of AI safety researchers, according to the source material; Block has not announced public access or released model weights or an API.
Outside coverage has described both natural-sounding moments and awkward ones. The Verge reported that some conversations felt “a little awkward and uncanny,” while finding Dessa more natural than another model featured in the comparisons. TechCrunch noted that it was unclear whether emotional delivery was being guided by tags. Block has not published a paper, model card or benchmark results to clarify how the system performs across broader tests.
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
Human-Like Voices and Fraud Risk
Dessa’s potential relevance goes beyond whether a demonstration sounds convincing. Voice impersonation fraud already exists, and more natural conversational timing could make it harder for someone on a call to rely on pauses, tone or emotional response as clues that a voice is synthetic. The source material cites the FBI’s 2025 Internet Crime Report, published in April 2026, which recorded 22,364 complaints mentioning AI and $893 million in reported losses. Those figures cover AI-related complaints and losses broadly; they do not establish that Dessa caused any fraud.
The same realism could also help voice systems respond more appropriately in customer service or other conversations. Block argues that many current systems fail to detect distress in a speaker’s voice. That is the company’s case for the research, not proof that Dessa reliably understands emotion. A model that produces emotional-sounding speech does not, on the evidence currently available, establish that it feels emotion or understands a person’s situation.
For businesses and families, the practical point is not that every familiar-sounding call is fake. It is that a voice alone should not verify a high-stakes request. Calls asking for urgent transfers, account access or sensitive information can be checked through a separately known number or another verification method. A pre-agreed family code word can add a further check when someone claims a relative is in trouble.
Products Worth Considering
Blueair Blue Pure 511i Max Air Purifier
Blueair Blue Pure 211i Max Air Purifier
LEVOIT Vital 100S-P Smart Air Purifier with Washable Pre-Filter
As an affiliate, we earn on qualifying purchases.
What Block Has Disclosed So Far
Dessa comes from a small research group at Block, the company behind Square and Cash App. The source material says the team was founded by former members of Dessa, a Toronto deep-learning startup acquired by Square in 2020. Block co-founder Jack Dorsey described speaking with the model as feeling like talking with a person, citing its emotion, pauses and cadence. The description reflects his reaction to the demonstrations, not an independent evaluation.
Block’s stated design differs from systems that convert speech to text, process the words and then generate a spoken reply. The company says Dessa is trained only on conversation audio and operates in a full-duplex manner, listening while it speaks. The source material also attributes to project lead Ragavan Thurairatnam the description of the work as “a GPT-3 moment for speech,” and says he characterized it as runnable on consumer GPUs. Block has not disclosed the model’s size or published benchmark results that would let others assess those claims.
The release has prompted sharply different reactions. Some listeners have praised the natural sound, while critics have questioned why voice AI should imitate people so closely. The dispute turns on the same feature: human-like speech may make interaction feel more natural, but could also reduce a listener’s ability to recognize a synthetic voice. The demonstrations show what Block chose to publish; they do not establish how Dessa performs in ordinary use or against attempts to deceive it.
Independent Tests Are Still Missing
As of October 26, no public technical paper, model card, benchmark or model weights had been identified in the source material. Block has not announced an API or public release date. Without outside access, it is not possible to verify the company’s claims about training, speed, conversational ability or performance across different speakers and situations.
Several practical questions are unresolved. Block has not listed the languages Dessa supports or explained how well it learns new ones. The source material also reports no public detail on watermarking, consent checks or whether the model can imitate a specific person. Block says Dessa chooses its own voice and personality, but that does not answer how the system would respond to a request to reproduce someone else’s voice or how misuse would be prevented.
It is also unclear how often the demonstrations’ emotional cues occur outside selected clips, how the system behaves in longer conversations, and whether its local hardware requirements will change. The available fraud statistics establish that AI-related crime and voice-cloning scams are concerns; they do not show that Dessa has been used in a scam.
Awaiting Researcher Access and Documentation
The next evidence that could clarify Dessa’s capabilities would be independent access, technical documentation and reproducible tests. Block has said a small group of AI safety researchers will receive early access, but the source material gives no schedule for their work or for publication of findings. A public release date has also not been announced.
For now, readers and organizations should distinguish between the model shown in Block’s recordings and a generally available product. Companies handling sensitive transactions can use verification methods that do not depend on recognizing a caller’s voice. Individuals can pause an urgent request, hang up and call a number they already trust, or use a family code word. These steps address existing impersonation risks; they do not depend on whether Dessa itself enters public use.
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 conversational cues such as laughter and pauses.
Can the public use Dessa now?
Not on the information available as of October 26, 2026. Block has offered a waitlist and early access for a small group of AI safety researchers, but has not announced a public release, API or model weights.
Has anyone independently verified Block’s claims?
No independent test is reported in the source material. Coverage has assessed Block’s recordings, but outside researchers have not publicly tested the model. Block has not released a paper, model card or benchmark results.
Does Dessa prove that AI can feel emotion?
No. The demonstrations may sound expressive, but emotional-sounding speech is not evidence that a model feels emotion. Whether Dessa understands the situations it responds to remains unestablished.
How can people check a suspicious voice call?
Do not rely on voice recognition alone for urgent requests involving money or sensitive information. Hang up and call a known number, verify through a separate channel, or use a family code word agreed in advance.
Source: ThorstenMeyerAI.com
