Aalto University: Donate Speech campaign awarded as the best European audio project in 2021
The Donate Speech campaign (Lahjoita puhetta in Finnish) won the best European Digital Audio Project of the Year 2021 at the PRIX EUROPA 2021 festival. PRIX EUROPA was founded by the European Parliament, the European Commission and the European Cultural Foundation in 1987. At the event, European television, radio and digital media productions are awarded.
The aim of the Donate Speech campaign is to collect all kinds of spoken Finnish as much as possible. The donated speech is used to develop speech recognition and artificial intelligence to better understand spoken Finnish. Taking part in the project are the Finnish Broadcasting Company Yle, the University of Helsinki, the Finnish State Development Company Vake (today the Finnish Climate Fund) and Aalto University, among others. Speech can be donated using a computer browser or a mobile app. Learn more about the campaign here (in Finnish).
The speech recognition research group at Aalto University is led by Professor Mikko Kurimo, who has been involved in the project from the start. Professor Kurimo says that Finnish and the majority of the world’s 6,000 languages suffer from the same problem: services that utilise speech recognition are not available, as using them requires fluency in English.
He believes that other small languages and countries can also model the successful and award-winning concept.
‘The collection has been a great success. There have been a lot of participants, technology has worked and the collected data has proven to be very important. As far as I know, there has never even been an attempt at this kind of fully virtual collection campaign led and marketed by a public television company, which also benefits the companies developing language technology, anywhere in the world’, Professor Kurimo says.
Now that over 4,000 hours of speech has been collected, Kurimo’s research group has a large task on their hands as they develop different automatic methods for examining, correcting, describing and classifying the material. So far, over 1,500 hours of the material has been transcribed, which has already helped to develop a better Finnish speech recognition system. The group has also created systems using the material that can identify the speaker’s age, gender and dialect.
‘The systems can be used to examine, for example, whether current artificial intelligence serves some groups of people better than others and how it could be made more equal’, Professor Kurimo explains.