Case: High quality speech to text in small languages

The Challenge

Infomedia and Podimo have struggled for a long time with speech-to-text conversion. Nordic Languages aren't well supported by major tech-companies, and the intensive manual task of transcribing TV, Radio and Podcasts is incredibly time-consuming, frustrating and usually has a lot of mistakes.

Our speech-to-text api supports 98 languages. Read more about that here.

What We Did

To tackle this problem, we not only needed a great speech-to-text model. We needed to create the best speech-to-text model in the world. All models on the market at the time created nonsensical transcriptions, with word error rates exceeding 20% (more than every 5th word was wrong). Instead, we combined our capacities on AI-modelling with a huge amount of local training data, getting the word-error-rate close to 5%, creating perfectly readable and understandable transcriptions in both Danish, Norwegian and Swedish.

How It Helped...

1: Infomedia With More Accurate Monitoring

Infomedia - scandinavia’s leading media monitoring company - uses our solution in their suite. It’s helped them reduce misspellings and monitoring mistakes on television and radio, leading to less work for the customer care service and happier customers.

2: Podimo with Better Recommendations and Happier Users

Podimo - a leading provider of exclusive podcasts - uses our speech-to-text solution to better understand all the content in their podcast. That way they can make better recommendations to their users based on the content of the podcasts, and not only the titles and genres.

What It Means Going Forward

With our speech-to-text solution, we’ve shown that converting speech to text in scandinavian languages doesn’t have to be difficult. We’ll keep on improving our solution so our customers can get even better results. And of course we also do speech to text in all the major languages.