Comparison
Dessa AI vs ElevenLabs: Native Emotion vs Voice Tags
ElevenLabs is the name most creators associate with expressive AI voice: give it a script, direct the delivery, and get polished audio back. Block's Dessa, announced October 23, 2026, approaches expression from the opposite direction — it was never given a script in its life. It learned emotion the way people do, by hearing it in real conversation. This comparison is therefore really about two philosophies: emotion as something a director specifies, versus emotion as something a model absorbs. One disclosure note for this money-page comparison: some links on this site may earn us a commission at no cost to you; today we link only to normal product homepages, and Block does not endorse either side of this comparison.
The fundamental difference: reading a script versus living a conversation
ElevenLabs is, at its core, a text-to-speech company. Its system starts from written words — your script — and renders them in a chosen voice. That makes it superb for produced audio: narration, audiobooks, video voiceovers, character lines and dubbing, where the words are known in advance and the job is to perform them well. ElevenLabs also offers conversational agent products, but its centre of gravity is performance of prepared text.
Dessa has no script and never will. As how Dessa AI works explains, Block trained it from scratch on human conversation audio only — no text, no pretrained models — so audio goes in and audio comes out, live. It is not performing your words; it is deciding its own words, timing and tone in real time, while listening to you. That is a categorically different product: a conversational partner, not a rendering engine.
Emotion: directed with tags versus learned natively
ElevenLabs' expressive control is famous for its audio tags: inline directions written into the script — cues such as [laughs], [whispers] or [sighs] — that tell the voice how to deliver a line. The director keeps control. If a laugh belongs after the second sentence, you put the tag there, regenerate, and it is there. For produced content that control is a genuine professional advantage: repeatable, editable and precise.
Dessa's emotion cannot be tagged, because there is no text layer to tag. Block says it natively hears and produces sarcasm, laughter, frustration and sadness because those signals were in its training conversations inseparably mixed with the words themselves. In its demos it laughs mid-sentence at the moment something is funny, gasps with what sounds like tearful gratitude, backchannels with "yeah!" and "got it!", and raises its volume when asked — a behaviour Block says nobody explicitly taught it. The trade-off is control: you cannot instruct Dessa's delivery line by line, because nobody writes its lines. Its expression is emergent, which is exactly what makes it both remarkable and, for production work, unusable as a rendering tool.
Live conversation versus produced audio
This is where the comparison resolves into use cases rather than a winner. Need a voiceover for a video today, in a consistent voice, with the emphasis exactly where your script puts it? ElevenLabs does that now, as a paid product with published pricing and an API developers can build on. Need an AI that holds an unscripted, interruptible, emotionally reactive conversation — finishing your sentence, reacting while you speak, adjusting in the moment? That is the behaviour Dessa was built to demonstrate, and the behaviour a script-first pipeline structurally struggles to fake, as we discussed in Dessa vs ChatGPT Voice.
Honesty requires the availability caveat in bold type: Dessa is a research preview. Block has released no product, no API, no pricing and no public release date, and published no benchmarks. Early access goes first to a small group of AI safety researchers, with a free public waitlist at block.xyz/dessa — see our waitlist guide. ElevenLabs, by contrast, is available to anyone today. A tool you can use beats a demo you can watch, for any deadline that matters.
Voices, cloning and responsibility
ElevenLabs built much of its business on voice choice and cloning — creating or replicating voices for production, a capability that carries well-known consent and impersonation duties. Block has not announced that Dessa can clone a named person's voice, and this site will not assume it can. The safety questions around a convincingly human voice are precisely why Block put safety researchers first, a rollout we examined in is Dessa AI safe? Both companies, in different ways, are selling or studying the same underlying fact: synthetic speech is now good enough that trust, consent and disclosure are product features, not footnotes.
Price and access, plainly
ElevenLabs publishes its pricing on its own site and sells subscriptions and API access today — check its official pages for current rates, which change. Dessa has no price because there is nothing to buy: no paid tier, no credits, no reseller programme. Anyone charging for Dessa access is not affiliated with Block. Block also says Dessa runs faster than real time locally on a MacBook Pro with an M4 Max chip or better and was trained on a single NVIDIA GB200 NVL72 — efficiency claims that, if they survive independent testing, would change the economics of live voice AI compared with per-minute cloud rendering. For now they are Block's claims, clearly labelled as such.
The verdict
If you make content, ElevenLabs is the working tool: scripted, tag-directed, repeatable emotion, available today. Dessa is the research direction: emotion learned natively from speech, expressed live and unscripted, unavailable today. The interesting question is whether the two converge — whether tomorrow's production tools absorb speech-native emotion, or whether live conversational models learn to take direction. Watch the tags-versus-native gap. It is one of the clearest fault lines in voice AI, and Dessa just made it impossible to ignore.
Sources: Block's official Dessa page (block.xyz/dessa) for Dessa's training, emotion and access claims; ElevenLabs' public product materials for its tag-based expressive text-to-speech, checked October 2026; The Verge and TechCrunch, October 2026.
Frequently asked questions
What is the main difference between Dessa AI and ElevenLabs?
ElevenLabs performs written scripts as audio, with emotion directed by inline tags such as [laughs] or [whispers]. Dessa holds live, unscripted conversations and produces emotion natively, learned from real speech — there is no script or tag layer.
Can Dessa AI replace ElevenLabs for voiceovers?
No. Dessa is a research preview with no product, API or pricing announced, and it does not render scripts. ElevenLabs is available today as a paid text-to-speech and voice platform for produced audio.
Which sounds more emotional, Dessa or ElevenLabs?
They do different things. ElevenLabs' emotion is directed and repeatable — you place the tag, it performs it. Dessa's emotion is emergent in live conversation, per Block's demos. No independent head-to-head evaluation exists.
🚀 Get Dessa launch alerts
Dessa is still a research preview. Join the free DessaAI Daily list and we will email you the day access opens — plus the week's biggest Dessa stories, once a week. No spam, no hype.
Form not loading? Subscribe on the signup page →