FCAI and Elisa make great strides in developing Finnish speech recognition applications

Researchers studied Finnish speech recognition and natural language processing using Elisa’s data. The outcome of this collaboration helps Elisa to offer better customer service.

Elisa customer service already makes use of the developed speech recognition technologies. Photo: Elisa Corporation

In 2019, the Finnish Center for Artificial Intelligence FCAI and Elisa, a Finnish pioneer in telecommunications and digital services, started an exciting research collaboration on Artificial Intelligence (AI) methods and Finnish language.

Researchers have studied Finnish speech recognition and natural language processing using data provided by Elisa. Now they have developed speech recognition technology that is more accurate in Finnish telephone conversations than the existing commercial technologies. Elisa announced recently that it will use the results to improve their customer service.

Developing Finnish speech recognition tools is very demanding, explains Mikko Kurimo, Associate Professor of Speech and Language Processing at Aalto University. “First of all, we have very little available conversational speech material and no standardized spelling for words of spoken language. In addition to the richness of spoken language, Finnish words can take several different forms, due to inflectional endings and compound words, and statistical modelling of the occurrence of these different forms is very difficult.”

FCAI and Elisa’s goal has been to improve the state-of-art of automatic speech recognition for Finnish language and enable development of better voice-based applications for end-users of Elisa. Researchers want to understand complex aspects of Finnish conversations and build speech recognition models that understand and analyze conversations automatically and transfer them from speech to text.

Another aim has been to build natural language processing models that would analyze unstructured text documents, such as emails. In addition, researchers want to provide tools to allow in-depth insight into corpora of real-world business documents. All this enhances the state-of-art of conversational AI applications.

The work conducted by FCAI and Elisa will lead to faster customer service, as customers can solve more issues for example with an AI-based machine, instead of waiting long times in a phone queue. AI applications can either solve problems independently or speed up the process when the customer needs to speak to a human expert. “Simple issues that are being handled constantly, such as change of address, could be done quite easily by using only a machine,” says Professor Kurimo.

Although not every single notification of change of address can be done with the assistance of a machine, by far the most of them could be handled in such a way. The rest of the contacts can be directed straight to a person who could help the client.

Elisa customer service already makes use of these speech recognition technologies and the first corporate clients of Elisa are piloting use of speech recognition in their own operations, in collaboration with Elisa.

Ville Rautio, CTO (Corporate Customers) at Elisa, explains that Elisa has a great amount of speech data that is of high quality and that has been collected from Elisa’s customer service and ICT service support. Access to these types of data makes development of well-functioning solutions possible. “We want to be a frontrunner in Finnish speech recognition because it is greatly beneficial for both Elisa and our customers,” he adds.

For Finnish language vitality, it is important to develop such AI applications in Finnish because it helps to ensure that Finnish customers get customer service in their own language also in the future. On the other hand, one of most central goals of FCAI is to ensure that Finland is a frontrunner in AI research. “For Finland, it would be good if we stayed on the frontline in these things. It is important to get more resources to develop the Finnish language technology,” says Kurimo.

Further information
Mikko Kurimo
Associate Professor, Speech and Language Processing
Aalto University, FCAI
Phone +358 50 3476221
mikko.kurimo@aalto.fi