Are you curious about getting to know the team behind Klang.ai? Today, Mattias Fält, Ph.D. in Mathematics and co-founder of Klang, answers Andreas' four quick questions about Klang and the technology behind the service.

How does a Ph.D. in Mathematical Optimization relate to AI?

That's a great question!

The foundation of all these AI models is based on mathematics. It's a collection of a lot of numbers in a large network that needs to be trained to get better at what it's supposed to do. Whether it's generating speech, text, or an image, everything is based on a lot of numbers.

In mathematical optimization, it's about finding the right numbers to make something as good as possible.

This fits well with what we're trying to do with these AI models. We're trying to make the model as good as possible, which means finding the right numbers.

What makes Klang's AI work so well for Swedish?

Klang's AI model is specifically trained to be good in Swedish, rather than being generally good at all languages. This has been achieved by training the model on a lot of sound and text in Swedish.

How do you train an AI model?

AI models are trained by showing them examples of sound and the corresponding transcribed text. By repeating this process millions of times, the model gets better at generating text from sound.

What is the biggest challenge in training AI models?

The biggest challenge is always to deliver high-quality transcriptions, even when the sound quality is poor, people talk over each other, or have difficult dialects.

Do you want to listen to the whole interview? Take a look at the interview in the service 👈

Did you know that this entire text was built with our functionality to find questions and answers from interviews? Try it yourself on Klang.ai ⭐