COLAI / CREATIVE THINKING FOR WHAT COMES NEXTINDEPENDENT IDEAS / COLIN MELVILLE
← AI RadarAI RADAR / 30 October 2026

Dessa: When AI Finds a Human Voice

Dessa's natural AI voices hint at a new era of conversation, characters and creative experiences. What happens when technology starts to sound human?

Two people pause mid-conversation across a café table, seen through a rain-marked window.

Every so often, something comes along that makes you think we're crossing another threshold.

Not because it's bigger, faster or more technically impressive.

Because something about it suddenly feels different.

This week, it was Dessa, a new AI voice model developed by Block.

I listened to the demonstrations and was genuinely taken aback by how natural it sounded.

Not polished. Not that familiar, slightly overenthusiastic AI voice we've all become accustomed to.

Natural.

The pauses. The interruptions. The subtle changes in tone. The sense that a conversation might actually be happening rather than a machine simply delivering its next response.

And that got me thinking.

We've been solving the wrong problem

For years, the development of synthetic voices has largely been about making speech more convincing.

Better pronunciation. More emotion. Greater realism.

And we've made extraordinary progress.

But human conversation isn't simply a sequence of beautifully delivered sentences.

It's unpredictable.

We interrupt each other. We hesitate. We laugh. We change direction halfway through a thought. Sometimes we communicate more through what we don't say than through what we do.

That messiness is part of what makes conversation human.

And it's incredibly difficult to reproduce.

Something interesting is happening

Dessa takes a different approach to voice interaction.

Rather than relying on the familiar process of transcribing speech into text, generating a response and then converting it back into audio, Block describes a system trained directly on human conversation.

It can reportedly listen and speak simultaneously, respond to interruptions and produce more natural conversational behaviour.

That's technically interesting.

But what interests me more is what it might enable.

Because once machines become genuinely conversational, the possibilities extend far beyond voice assistants.

From interfaces to experiences

Imagine interacting with a fictional character who doesn't simply follow a predetermined script.

Someone who remembers previous conversations, responds to your personality and develops an ongoing relationship with you.

Imagine entertainment that changes depending on who's experiencing it.

Or a brand relationship that feels less like navigating a customer-service system and more like interacting with a recognisable personality.

Think about education, gaming, live experiences, music, storytelling and entirely new forms of entertainment.

Not simply technology responding to instructions.

Technology participating in experiences.

That's where my imagination starts running.

And it raises a much bigger question about what we mean by an interface.

Perhaps the next generation of digital experiences won't feel like interfaces at all.

Realism isn't the destination

There's an important caveat.

Dessa is still a research preview. The demonstrations are impressive, but they don't establish how reliably the system performs in real-world conditions.

And sounding human isn't the same as understanding a human being.

There are serious questions around trust, consent, identity and the emotional relationships people might form with increasingly convincing synthetic personalities.

Those questions matter just as much as the technology itself.

But I don't think we should underestimate the creative possibilities.

For me, the most interesting technological developments aren't necessarily the ones that solve an existing problem.

They're the ones that make you start imagining things you couldn't previously build.

Dessa feels like one of those moments.

We've spent years teaching machines to generate convincing speech.

Now we're beginning to explore what happens when they can participate in the conversation.

And I suspect that's where things get really interesting.

Editorial disclosure: Dessa was announced by Block on 23 October 2026. These observations are based on its publicly available demonstrations and published technical information, rather than independent hands-on testing.

I’m Colin Melville, an Executive Creative Director. These views are my own.

More from AI Radar ↗