Threll.ai
Technology and automation

Voice AI: The forgotten revolution – and the challenges we need to solve

Language models like ChatGPT have made the whole world sit up and take notice. We see how text-based AI is changing the way we work. But what about voice? Why does Voice AI still feel like an unfinished technology? In this blog post, we show why Voice AI is about to become the next big step – and what challenges we need to solve to get there.

BySveinung Totland6 min read
A woman in a beige blazer presents in front of a board illustrated with a microphone, a head and a sound wave, while two colleagues listen

Part 1: Voice AI – The Forgotten Revolution

Most people have now become aware of what language models can do. ChatGPT and other AI assistants have evolved from technological curiosities to tools that actually help people in their daily work routines. It is no longer science fiction to have a conversation with an AI that can write contracts, explain quantum physics, or give strategic advice at the level of McKinsey. We see it. We write with them. We understand that this will change everything.

But there is one thing we do not do: We do not talk to them.

Why is it so quiet around Voice AI?

It's actually strange. Because at the core, it's the same technology driving both text-based and voice-based AI systems. So why does conversing with text feel smart, while talking with voice still feels... stupid?

But what if it has changed now?

Because it has. Voice AI has made significant technological leaps over the past year. New models for text-to-speech, speech recognition, and conversation management enable the creation of conversations that actually feel human – and meaningful. Not only technically impressive, but emotionally intuitive.

At Threll.ai, we have chosen to focus on exactly this. Not because it is "cool," but because we believe voice will become the next dominant way we communicate with AI. Not just because it's possible – but because it is natural.

Speech is humanity's most advanced interface

Language – especially speech – is our most effective form of communication. It is faster than text. It carries more context, emotions, and intention. It is the first thing we learn as children, and the last we lose in old age. Speech is both intuitive and universal.

So why shouldn't this be our preferred way to collaborate with AI?

From conversation to workforce

It's not just about talking. It's about workflow. With modern Voice AI – like our Voice Thrells Snorre (inbound) and Harald (outbound) – you don't just get a chat. You get:

  • A voice that remembers what you've said (📚 memory)
  • A voice that's connected to your systems (🔗 integrations)
  • A voice that follows up, documents, and improves itself (⚙️ workflow)

And that's exactly where the difference lies. Voice AI is not valuable on its own – it needs to be connected to what's important in your business. CRM, calendars, support systems, databases. Triggers, notifications, call history, insights.

Without this, Voice AI risks only becoming a pleasant conversation – without any real business value.

That's why we work closely with the customer's APIs and integrations at Threll.ai. We know that magic happens when the conversation gains the power to act – when a Voice Threll not only understands you, but acts on your behalf.

Unfair competition

Voice AI compensates for human weaknesses – and enhances strengths:

  • A sales representative doesn't need to remember all the details – Threllen does it.
  • A customer service agent doesn't need to be at their best every day – Threllen always is.
  • A manager doesn't need to listen to 50 conversations – Threllen transcribes, analyzes, and provides insights automatically.

It's like having an entire department in your ear. And it's never sick, never cranky, and always on the job.

From Chat to Talk

It is also worth considering the name: ChatGPT. What does it really mean to chat? Sending text messages? Not necessarily. The word "chat" comes from the English word for "talk." Talking.

So why are we sitting here writing with AI – when we could actually be talking?

Want to talk it through with someone who has done this before?

Fifteen minutes on a call. We will tell you honestly whether a voice agent fits your setup – and what has to be settled before it goes live.

Book Meeting (opens in a new window)+47 24 12 66 30

15 minutes · no obligation · pick your own time

This is the start of something big

At Threll.ai, we believe Voice AI is the next big step in the AI revolution. Not because it's fancy, but because it's right. Right in form. Right in function. And right for the future.

Because while the world is still writing – some of us are starting to speak.

And we get answers.

Part 2: Voice AI – The Revolution with Challenges

Speech is beautiful – and difficult

We humans understand each other despite mistakes. We mumble, interrupt, speak in codes. But we fill in the gaps automatically. For an AI, this is extremely challenging.

It is still difficult to:

  • Keep the thread in a conversation over time
  • Understand context, intent, and subtext
  • Adapt to different speaking styles, dialects, and pauses

Even the best Voice AI systems can lose their footing in a complex conversation – and then the illusion quickly breaks.

Siri syndrome

Most people have spoken with a voice assistant before – and many have been disappointed. Waiting times, strange responses, lack of understanding. The experience has left a mark on their memory:

"Voice AI is slow and stupid."

This is perhaps the biggest barrier Voice AI faces today: low expectations. People have given up before they've even started. To succeed, we must not only build better technology – we must rebuild trust.

Real-time = hard work

Speech requires real time. This means that multiple AI components must work simultaneously and seamlessly:

  • Speech → text (ASR)
  • Text → understanding (NLU)
  • Understanding → response
  • Response → speech (TTS)

All of this must happen without noticeable delays. It requires a lot of computing power, optimization, and precision. Compared to text-based AI, it's a high-speed logistical dance – and a technical nightmare if you misstep.

Integration or isolation

Voice AI is not valuable on its own. It's when it connects to your systems that the magic happens:

  • CRM, calendars, databases, support systems
  • Triggers, notifications, follow-up
  • Memory, learning, and automatic improvement

But many companies have closed infrastructure, and integration takes time. Without this, there's a risk that Voice AI will just become a fancy voice chatbot – not a real employee.

Ethics, privacy, and trust

Voice data is sensitive. You hear everything: tone, emotions, sometimes full names and bank account numbers. This requires:

  • Transparent data processor agreements
  • Ability to anonymize
  • Control over what is stored and how it is used

And perhaps most importantly: You must be able to trust that the voice agent does not cross the line – technologically, legally, or humanly.

People are not ready – but they will be

Perhaps the biggest challenge is cultural. People like to chat, but many hesitate to speak. They are unsure how to phrase themselves. Or they fear it might "sound stupid." But that changes. And when it does, it changes quickly. We see the same pattern as when text-based AI took off:

  • First curiosity
  • Then fear
  • Then astonishment
  • And finally: adoption

Conclusion: The revolution comes in the form of a voice

Voice AI is not magic. It is hard work, system integration, security procedures, and innovation in user experience. But it is also a doorway to a new way of collaborating with technology – where your voice controls, the AI listens, and the conversation becomes value-creating in real time.

At Threll.ai, we believe not only that this is possible.

We are building it right now.

Frequently asked questions

That experience is the barrier itself, not an argument against trying again: the models for text-to-speech, speech recognition and conversation management have taken large leaps over the past year, while people's expectations are still stuck where they were. So plan on rebuilding trust, not just buying better technology. Start with one call type where you can hear for yourself whether the illusion holds, and let a good experience convince the rest of the organization instead of a slide.

No, but then pick a use case where the value sits in the conversation itself. An agent that answers, transcribes, analyzes and documents gives you insight without writing anything back into your systems. What you should not do is promise actions the agent cannot carry out — without a connection to CRM, calendars and support systems it stays a pleasant conversation with no real business value. Treat the integrations as phase two and plan them from day one.

Three things belong on paper before the first call rather than after: the data processing agreement, how the sensitive parts are anonymized, and what is actually stored and for how long. Voice data carries more than the words, since tone and emotion come along with the recording, so "we keep everything just in case" is rarely a position you want to defend later. Work it through with legal and your processor while you are still choosing a vendor.

Not necessarily, but this is exactly where even good systems lose their footing, and the illusion breaks fast once the thread is gone. So design for the break instead of hoping it won't come: keep the task narrow enough that the agent knows what the call is about, and make sure there is a clean handover to a human when it doesn't. A system that admits it did not understand costs you less trust than one that guesses.

It is more than a layer on top. Speech requires recognition, understanding, response generation and synthesis to run at the same time with no noticeable delay, and a pause that is perfectly fine in a chat window reads as a dead line on a phone call. The real-time requirement, not the language model, is the heavy part — expect work on computing power, optimization and precision rather than one more button in the interface.

Related Articles

Continue reading more articles

See all articles