- The aim is to ensure that speakers of less widely spoken languages have access to high-quality speech-based AI services and products, which are not necessarily developed by large global companies.
- The project will develop speech recognition and synthesis for Finnish, Finland-Swedish and the Sámi languages. The solutions will be tested in, for example, phone services and translation. Collaboration partners include companies, public administration and associates from Norway, Sweden and Estonia.
- The results will include commercialisable models based on which speech interfaces can be developed for small languages and dialects. They will help Finnish companies with AI solutions compete in the international market.
ChatGPT and other AI applications are rapidly gaining traction. They can support many types of work, including customer service and healthcare. However, we are facing a major transformation: in the future, people will increasingly communicate verbally with AI. This will facilitate the everyday use of AI applications.
Consequently, it is important to develop speech interfaces in languages with fewer speakers, such as Finnish, Finland-Swedish and the Sámi languages. The University of Helsinki, Aalto University and associates are working to do just this in the Business Finland–funded
The researchers are creating replicable models to help companies develop speech-based AI applications. This will pave the way for international export, as the voice recognition market, for example, is expected to
“Large language areas are home to a variety of dialects. Products based on AI should at least be able to understand them even if responding in the majority language,” says Research Director Krister Lindén of the Department of Digital Humanities.
LAREINA is connected to
“It would free up time for patient care,” notes Lindén.
AI could also provide conference interpreting services, complete time-consuming transcription work and handle switchboard duties. During the project, various applications will indeed be tested both in companies and in the public sector.
Whereas Aalto University is responsible for developing speech recognition and language models, the University of Helsinki as the project coordinator provides expertise in speech synthesis. This is required to ensure AI can talk to users. The aims are to teach AI to talk with as small a dataset as possible and to achieve a natural-sounding result.
“Important issues can be weighted in sentences in different ways, and breaks added where appropriate.”
AI is being trained with speech data from the Finnish
Our goal is to create a speech interface with as small a dataset as possible. This can be done in various ways. For example, we can collect data in related languages and then hone the data in the target language.
The language models created in the project will be made openly available with a licence enabling commercial use. They will be published in the AI community
“It also benefits society, which is important in a publicly funded project,” notes Tietoevry’s Head of R&D Iftikhar Ahmad.
Tietoevry is keen to ensure that its AI applications provide added value and the company obtains valuable information important for its operations. This is where collaboration with universities can be useful. Ahmad believes the project could even be used to develop a Nordic equivalent to the AI solution
The ecosystem and infrastructure developed in LAREINA will support the participation of Finns in EU projects. Ahmad believes it would be important to obtain more European funding to boost export and growth-promoting networks.
“It’s a huge opportunity for Finnish researchers and companies.”
Another partner interested in the potential of speech technologies is the
“We have high expectations and hopes for this project. It’s great to be involved.”
Please contact us and we will tailor a project according to your individual needs: