Threll.ai
AI telephony

Should future Voice AI be built here at home – or along American lines?

Everyone is talking about chatbots and text-based AI. But the real revolution – the one that will change customer dialogue, service, and value chains – is happening right now, in the phone.

ByKjell Rune Monsø3 min read
A meeting room where several attendees are asleep at the table while a woman presents a poster of a telephone handset wired to a brain marked with the Norwegian, American and Chinese flags

For the first time, a company can respond to every customer, every time – with an AI that speaks Norwegian, Swedish, and Danish, understands context, and acts in real time.

This is AI telephony – and it will change everything. The question is just: Should Scandinavia own the voice itself – or let others do it for us?

When AI awakens – telecommunications must keep up

The Scandinavian telecom industry is world-leading in coverage, fiber, and infrastructure. But in the field of intelligent telephony – voices that actually understand and act – we are behind.

Today, nearly all AI telephony in Norway and Scandinavia is routed through American lines like Twilio and Telnyx. This means that audio, conversations, and customer data are sent out of the region every single day.

Not because the technology doesn't exist here – but because no one has connected telecommunications and AI in the same ecosystem.

This is about more than telephony

This is about digital sovereignty, data security, and the experience of intelligence.

When the voice is sent out of the country, data, insights, and innovation opportunities are also sent out. When the delay between speech and response exceeds 100 milliseconds, the human rhythm in the conversation disappears.

In short:

  • The further the signal travels, the less human the AI sounds.
  • The further the data center is, the more vulnerable the conversation becomes.

Threll.ai: The engine is looking for the pipes

Threll.ai has developed the conversation engine – the AI that speaks, understands, and acts in Scandinavian languages. It is trained on real conversations, Scandinavian culture, and Norwegian dialect. It works.

But we need those who own the pipes. We need telecom partners who can deliver SIP trunk, HD Voice, and local data centers in Oslo, Stockholm, and Copenhagen – with low latency and WebSocket support for AI integration.

This is not science fiction. This is standard technology. But the infrastructure must exist – here at home.

A Scandinavian effort

This is our ambition:

Building the world's most natural voice AI – in Norwegian, Swedish, and Danish – and doing it on Scandinavian infrastructure.

We seek partnerships with telecom players who want to own the future voice economy, not just transmit it. Because at the intersection of networks and intelligence lies a new industry ready to emerge.

Norwegian network. Scandinavian AI. Local control.

The revolution has begun. The question is not whether you want to join – but whether you want to lead.

Contact us – let's build the Scandinavian voice before someone else does.

Written by Threll.ai

We are building Scandinavian voice intelligence – technology that understands, responds, and acts. Voices that convert.

Frequently asked questions

Not overnight, but you should know what the route costs you. As long as calls travel over American lines, audio, conversation content and customer data leave the region every single day, and part of your latency budget is spent on distance rather than on understanding the caller. Turn that into two measurable requirements, where the data actually sits and how long the delay actually is, and decide on those numbers rather than on the vendor's name.

It means two layers at once. The network and the trunk are Norwegian in origin, while the AI and the languages are Scandinavian, with local data centers in Oslo, Stockholm and Copenhagen. So the question to put to a supplier is not whether something is Norwegian, but which leg terminates where, since signaling, the media stream and the model itself can sit in three different places.

Yes, but measure it mouth to ear across the whole path, not as response time inside the model. Above roughly 100 milliseconds the human rhythm of the conversation disappears, and on a route out of the region much of that budget is gone before the AI has begun to answer. Ask to be shown where the media stream is actually terminated, and test with real callers on mobile rather than only from the office network.

The physics does, the language coverage does not. The point that a longer signal path sounds less human and that a distant data center makes the conversation more fragile holds wherever you sit, so the same two questions about latency and data location are worth asking your own provider. What is specific here is the engine itself, which speaks Norwegian, Swedish and Danish and is trained on Scandinavian conversations and Norwegian dialect.

No, but write the requirements now. The conversation engine already exists and works, and what is missing is not new technology but telecom partners delivering SIP trunk, HD Voice, low latency and WebSocket support locally. If those four items are in your procurement today, you can move the traffic home the day the pipes are there instead of rebuilding the whole setup.

Related Articles

Continue reading more articles

See all articles