Comparison

Dessa AI vs Hume EVI: Two Different Bets on Emotional Voice

2026-11-04 · 7 min read · By DessaAI Daily

Most voice AI comparisons are about speed or price. Dessa AI vs Hume EVI is about something narrower and more interesting: emotion. Both systems claim a machine should hear how you feel, not just what you said. They go about it in almost opposite ways — and only one of them can be used by a developer today.

Dessa is Block's full-duplex, speech-to-speech model, announced October 23, 2026 as a research preview. Hume's Empathic Voice Interface, known as EVI, is a developer API from Hume AI. This comparison uses only verified facts about Dessa from Block's official page, and Hume's own description of EVI. Where Block has not announced something, we say so plainly.

How each system is built

Dessa was trained from scratch on human conversation audio only — no text, no pretrained models. Block says it learned language from hundreds of times fewer words than text large language models need, was trained on a single NVIDIA GB200 NVL72 system, and runs faster than real time locally on a MacBook Pro with an M4 Max chip or better. Everything it knows about language, timing and feeling came from listening, as we explained in how Dessa AI works.

Hume describes EVI differently. On Hume's own blog, EVI is powered by an empathic large language model that processes your tone of voice, uses that tone for end-of-turn detection, and generates responses while modulating its own tune, rhythm and timbre. Developers connect over a WebSocket, streaming audio in and receiving responses in real time, and Hume says EVI can be configured with other language models for longer responses. In other words, Hume builds an emotional layer around language-model technology. Dessa removes the text language model entirely and asks whether emotion and language can be learned from sound alone.

Emotion: native learning vs measured prosody

Block's claim for Dessa is that emotion is not a feature added later. Because its training data was real conversation, sarcasm, laughter, sighs and frustration arrived together with the words, in context. In Block's demos Dessa laughs mid-sentence, backchannels with “yeah” and “got it”, gasps with what sounds like tearful gratitude, and adjusts its volume when asked — a behaviour Block says nobody explicitly taught it.

Hume's claim is more measurable in spirit. Hume says EVI provides streaming measurements of the tune, rhythm and timbre of a user's speech using its prosody model, integrated with its empathic model. That framing will appeal to developers who want a signal they can log, inspect and build product logic around. Dessa offers no such published measurement, no benchmark scores and no parameter count — so there is no honest head-to-head score to report here.

That difference matters for trust. A developer choosing EVI can read Hume's documentation, open an account and test the API against their own callers. A researcher watching Dessa can listen to Block's selected demo recordings and wait for the first independent evaluations from the safety researchers who get access first.

Tools, knowledge and getting work done

Here the practical gap is wide, and it favours the product that exists. Hume says developers can customise EVI — the system prompt, the language model and the tools EVI can use — through its configuration work. A voice interface designed to be paired with other models and tools is, by design, aimed at applications: support, coaching and other live conversations where looking something up matters.

Dessa, by contrast, has never read a book, a manual or a webpage, by design. Block has published no evidence that it can call tools, browse, or handle precise factual work reliably, and we examined those limits directly in what Dessa AI can't do yet. That is not a hidden flaw; it is the research question Dessa poses. Can conversation alone produce useful knowledge and reasoning, the way listening does for a child? The demos show fluency and social timing. They do not yet show checkable, tool-using assistant work.

Access and price: the deciding factor today

Hume EVI is an API product a developer can sign up for and integrate today, on Hume's published terms — check Hume's own site for current pricing before you budget, because API pricing changes and this article does not quote a rate we did not verify on Hume's official pricing page today.

Dessa has no price at all, because there is nothing to buy. Block has announced no pricing, no API, no weights and no public release date. Joining the waitlist at block.xyz/dessa is free, and early access goes first to a small group of AI safety researchers, as our Dessa AI waitlist guide explains. Anyone selling Dessa access today is not Block.

Which should you choose?

Choose Hume EVI if you are building now and you need an emotionally aware voice interface you can actually integrate, configure and test with real users. Choose to watch Dessa if you care about the scientific bet underneath it: that a model trained only on speech might handle timing, interruption and feeling more naturally than any system that translates feeling into text and back again.

The fairest summary is asymmetrical. EVI is a tool. Dessa is an experiment with a waitlist. If Dessa's speech-only claim holds up under independent testing, future comparisons will be about quality. Today, the honest comparison is about availability — and on availability, only one of these two is open for business.

Sources: Block's official Dessa page (block.xyz/dessa); Hume AI's official EVI announcement on hume.ai; The Verge and TechCrunch coverage, October 2026.

Frequently asked questions

← More Dessa AI articles Join the official waitlist ↗