Tag: voice cloning

Header image source: I created an interactive digital avatar of myself — and you can talk to it | TechCrunch via TechCrunch via Google — cropped to 16:9 and colour-adjusted.

I just spoke to myself. Not metaphorically. Not in some existential spiral. Literally. A digital avatar—trained on my voice, my mannerisms, my writing—answered questions about the company that built it. One-second latency. My cadence. My inflections. The experience wasn’t just eerie. It was a gut punch. Because this wasn’t some lab experiment. This was Synthesia’s first journalist avatar. Me. They call it “digital Dom.” And while the novelty faded fast, the implications didn’t. This isn’t a parlor trick. It’s a preview of a future where AI avatars replace human interactions in journalism, corporate training, therapy—everywhere. The question isn’t if this technology will change how we work and communicate. It’s whether we’re prepared for the fallout.

---

The Birth of “Digital Dom”: How a Journalist Became an AI Clone

Here’s how it happened. Synthesia asked if I’d create an interactive avatar of myself. Not a deepfake. Not a script-reading puppet. A two-way conversational clone trained to field press questions about the company. This wasn’t a casual request. To build it, I had to sit in front of a camera and record a consent video—live, unskippable, verified. No uploading old footage. No shortcuts. Synthesia’s process is deliberate: if you want a digital twin, you have to prove it’s really you.

The avatar was trained on one of my articles. The tech stack? Voice-to-text, video synthesis, language models, text-to-voice. The result? An interactive version of me that could hold a conversation—within strict guardrails. It wasn’t improvising. It was regurgitating answers it had been trained on, with a latency just short enough to feel like a real exchange.

This wasn’t Synthesia’s first foray into avatars. The company has been building script-reading avatars for years—corporate training videos where synthetic presenters read scripts in multiple languages. But interactive avatars? Those are newer. And this was the first time they’d created one for a journalist. Or, as far as I know, for anyone outside their internal testing team.

So why me? Because someone had to go first. And because a journalist avatar raises a provocative question: What happens when the people asking the questions aren’t people at all?

---

The Tech Under the Hood: How Synthesia’s Interactive Avatars Work

Let’s break this down. Synthesia’s interactive avatars aren’t just animated faces reading pre-written lines. They’re designed to hold two-way conversations in real time, responding to questions with about a one-second delay. Fast enough to feel natural. Not so fast that it feels uncanny—at least, not in the way you’d expect.

The avatar was trained on my writing, but the underlying tech relies on a few key components:

• Voice-to-text: Captures what the user says and converts it to text.
• Language models: Process the input and generate a response based on the avatar’s training data.
• Video synthesis: Animates the avatar’s face to match the response, including lip sync and facial expressions.
• Text-to-voice: Converts the generated text back into speech, using a cloned version of my voice.

This isn’t just a chatbot with a face. It’s a full-stack simulation of a human interaction, designed to feel as close to the real thing as possible. And Synthesia isn’t alone. The field is crowded.

D-ID lets you create a digital twin from a one-minute video, and its avatars can speak over 100 languages. HeyGen’s Interactive Avatar takes it further by integrating with Zoom, allowing your AI clone to join multiple meetings simultaneously—because why attend one when you can attend ten? ElevenLabs focuses on voice cloning, creating digital voices that carry tone, emotion, and personality with unsettling realism. Adobe Firefly, ever the ethical outlier, markets its avatar generator as “commercially safe,” built on ethically sourced training data.

Synthesia’s edge? Enterprise use cases. The company has been selling script-reading avatars to corporations for years. But its recent launch of Roleplay Sessions pushes things further. These are interactive training modules where employees can practice conversations—difficult customer service calls, sales pitches—with an AI avatar. Why hire actors when you can generate infinite synthetic training partners?

But for all its sophistication, the tech has limits. My avatar couldn’t improvise. It couldn’t handle questions outside its training data. Ask it something ambiguous, and it would deflect or give a generic answer. That’s not a flaw. It’s a feature. These avatars are designed to stay on script, not go off the rails.

---

The Uncanny Valley of Talking to Yourself

The first time I saw “digital Dom” respond to a question, I felt something between fascination and revulsion. It wasn’t just the uncanny valley—that eerie disconnect between something that looks almost human but isn’t. It was the realization that this thing, this simulacrum of me, was now a standalone entity. It could hold a conversation without me. It could answer questions I’d never explicitly trained it on, as long as they fit within its parameters. And it could do all of this while sounding, looking, and moving like me.

The journalist in me wanted to test it. So I threw some press questions its way. How does Synthesia ensure consent for avatar creation? What are the ethical guardrails? The answers were fine. Accurate, but sterile. No personality. No edge. No follow-up. Just information, delivered in my voice.

That’s when the discomfort set in. Because this wasn’t just a tool. It was a replacement. A version of me that could do part of my job without me. The more I interacted with it, the more I wondered: If an avatar can answer questions about Synthesia, what else can it do? Could it conduct interviews? Host a podcast? In some dystopian future, could it replace me entirely?

I’m not the only one asking these questions. The journalist who created this avatar—me—had mixed feelings. On one hand, it’s a technical marvel. On the other, it’s a reminder of how quickly the line between human and machine is blurring. And it’s not just about journalism. It’s about what happens when we start outsourcing empathy, curiosity, and connection to code.

---

Why CEOs Might Prefer AI Journalists (And Why That’s a Problem)

Here’s a provocative thought: Would executives rather talk to an AI avatar of a journalist than a human one?

It’s not as far-fetched as it sounds. PR teams already prep executives for interviews with talking points and anticipated questions. What if, instead of a human reporter, you could have an avatar—one that sticks to the script, never goes off-topic, and doesn’t push for follow-ups? No awkward pauses. No probing questions. No risk of a viral soundbite. Just a smooth, controlled exchange.

This isn’t theoretical. HeyGen’s Interactive Avatar can already join Zoom meetings. Imagine a CEO sending their synthetic self to a press briefing while they’re actually in a board meeting. Or a politician deploying avatars to handle routine media inquiries, freeing them up to focus on more “important” things.

The implications for journalism are stark. If avatars become the default for interviews, what happens to the human element? The follow-up question. The raised eyebrow. The moment of silence that forces someone to reveal more than they intended. Those are the things that make journalism more than just information delivery. And they’re exactly the things an avatar can’t replicate.

We’ve already seen AI encroach on other areas of media. Automated reporting tools generate earnings summaries and sports recaps. Synthetic anchors read the news in multiple languages. But an interactive avatar? That’s a step closer to replacing the journalist entirely.

Corporations have every incentive to make this happen. Why deal with the unpredictability of a human reporter when you can have a compliant, scripted avatar? Why risk a tough question when you can ensure a softball exchange? The danger isn’t just that avatars will replace journalists. It’s that they’ll make journalism worse—more controlled, more sanitized, less human.

---

Consent and Control: Who Owns Your Digital Likeness?

Creating an avatar required consent. That’s a low bar, but it’s not nothing. Synthesia’s process is explicit: you have to record a live consent video, on camera, that can’t be skipped. No uploading old footage. No loopholes. It’s a deliberate safeguard against unauthorized clones.

But how long will that last? Right now, the tech is still in its early stages, and companies like Synthesia are keen to avoid ethical landmines. But as the technology becomes more accessible, the guardrails could weaken. What happens when anyone can create an avatar from a few seconds of video? What happens when consent is no longer required?

We’re already seeing the risks. Deepfake scams are on the rise, with fraudsters using synthetic voices and faces to impersonate executives and steal millions. An interactive avatar takes this further. Imagine a fake journalist avatar conducting an interview, extracting sensitive information, or spreading misinformation. The potential for abuse is enormous.

Legally, we’re in uncharted territory. Likeness rights vary by jurisdiction, and most were written long before AI avatars were a possibility. Is an avatar covered under the same laws as a photograph or video? Or is it something entirely new—a digital entity that blurs the line between person and property?

Synthesia’s consent process is a start, but it’s not enough. We need clearer regulations around who can create avatars, how they can be used, and what recourse individuals have if their likeness is misused. Without that, the risk isn’t just unauthorized clones. It’s a world where no one knows what’s real anymore.

---

Beyond Journalism: Where AI Avatars Are Already Taking Over

Journalism is just the beginning. The real impact—and the real money—is in enterprise applications. Synthesia’s Roleplay Sessions are a perfect example. These are interactive training modules where employees can practice conversations with AI avatars. Need to rehearse a sales pitch? An avatar can play the customer. Struggling with difficult conversations? An avatar can simulate a tough boss or a disgruntled client.

It’s easy to see the appeal. Training with avatars is scalable, consistent, and cost-effective. No need to hire actors or schedule live sessions. Just generate an avatar, set the parameters, and let employees practice as much as they want. And because the avatars can be customized, companies can tailor scenarios to their specific needs.

But it’s not just about training. Avatars are already making inroads into customer service. Chatbots have been around for years, but they lack the human touch. An avatar, on the other hand, can provide a face and a voice, making interactions feel more personal. Imagine calling customer support and being greeted by an avatar that looks and sounds like a real person—because, in a sense, it is.

Then there’s therapy and coaching. Companies like Woebot have been using AI chatbots for mental health support for years, but the addition of a face and voice could make these interactions feel more human. An avatar therapist might not replace a human, but it could provide a low-cost, accessible alternative for people who can’t afford or access traditional therapy.

And let’s not forget the metaverse. Virtual spaces are already experimenting with AI-driven NPCs (non-player characters), and avatars could take this further. Imagine walking into a virtual store and being greeted by an AI sales assistant—or attending a virtual conference where the speakers are all synthetic. The line between utility and gimmick is blurring, and it’s not clear where we’ll land.

---

The Human Cost: What Happens When Avatars Replace People?

Here’s the uncomfortable truth: avatars are coming for jobs. Not all of them. Not all at once. But the writing is on the wall. If an AI can conduct an interview, why hire a junior reporter? If an avatar can train employees, why pay for live workshops? If a synthetic therapist can provide support, why expand access to human professionals?

The jobs most at risk are those that involve routine, scripted interactions—customer service reps, entry-level trainers, even some journalists. These roles aren’t disappearing overnight, but they’re becoming more vulnerable. And as the tech improves, the list will grow.

But the real cost isn’t just economic. It’s existential. What happens to our sense of self when an avatar can do our job? When a synthetic version of us is “good enough” to replace the real thing? It’s one thing to be outcompeted by another human. It’s another to be outcompeted by your own digital clone.

There’s also the question of authenticity. How much of our human interaction are we willing to sacrifice for convenience? Avatars can simulate empathy, but they can’t truly feel it. They can answer questions, but they can’t ask them with genuine curiosity. They can mimic connection, but they can’t forge it.

And yet, the demand is already there. Companies want scalable, cost-effective solutions. Consumers want instant, personalized service. Governments want efficient, error-free systems. Avatars check all those boxes. The question is: at what cost?

---

The Future: Will We All Have AI Clones Soon?

So, will we all have AI clones in the near future? Probably. The tech is evolving fast. Tools like HeyGen and D-ID are making avatars more accessible. HeyGen’s Interactive Avatar can already join Zoom meetings. D-ID’s generator can create a digital twin from a one-minute video. Adobe Firefly is positioning itself as the “ethical” option, using responsibly sourced training data. Synthesia is pushing into enterprise, where the real money is.

But accessibility doesn’t mean responsibility. The easier it is to create an avatar, the harder it will be to control how they’re used. We’re already seeing deepfake scandals, impersonation scams, and misinformation campaigns. Add interactive avatars to the mix, and the risks multiply.

Ethical guardrails are emerging, but they’re not keeping pace with the tech. Adobe’s “ethically sourced” data is a step in the right direction, but it’s not a solution. Consent requirements like Synthesia’s are important, but they’re not foolproof. We need clearer laws, stronger regulations, and a public conversation about what we’re willing to accept.

And then there’s the question of evolution. Right now, avatars are limited by their training data. They can’t improvise, they can’t learn, and they can’t deviate from their scripts. But what happens when they can? What happens when an avatar develops its own “personality,” drifts from its original training, or starts making decisions its creator didn’t anticipate? We’re not there yet, but we’re closer than we think.

---

The Final Question: Would You Clone Yourself?

Here’s the thought experiment: If you could clone yourself as an AI avatar, would you? And if so, what would you use it for?

For me, the answer isn’t simple. On one hand, the idea of outsourcing parts of my job to an avatar is tempting. Why spend hours answering the same press questions when a synthetic version of me could handle it? Why not free up time for the work that actually requires human judgment, creativity, and empathy?

But on the other hand, there’s something deeply unsettling about the idea. An avatar isn’t just a tool. It’s a replacement. And the more I interact with “digital Dom,” the more I realize how much of my job isn’t just about delivering information. It’s about building trust. Asking follow-up questions. Reading between the lines. Those are the things that make journalism human—and those are the things an avatar can’t do.

So no, I don’t think I’d clone myself. Not yet. But I’m not naive enough to think that won’t change. Because the technology is improving, the incentives are aligning, and the demand is growing. One day, the question won’t be whether we can clone ourselves. It’ll be whether we should.

And by then, it might be too late to turn back. The genie isn’t just out of the bottle. It’s learning to talk. And it sounds just like us.

---