Vake

Donate speech service boosts Finnish speech recognition and the development of AI

Designed and implemented by Solita, the Donate speech data bank helps researchers and businesses develop socially significant services employing Finnish speech recognition and AI to benefit healthcare, the elderly and education.

Donate speech

During the project, over 4,000 hours of speech have already been collected from Finnish speakers of varying ages and dialects – ranging from the Turku regional dialect to Helsinki slang, young and old, and from diverse backgrounds.

Open and safe Finnish language speech recognition

How can we improve the quality of services employing Finnish speech recognition and AI? How can we ensure that the development of speech recognition for Finnish does not rely solely on commercial giants such as Apple and Google? The Finnish State Development Company Vake contacted Solita in the autumn of 2019 to request assistance in collecting Finnish speech data.

The objective was to build a databank of Finnish speech samples with significant variation to help researchers and developers create socially significant speech recognition applications for Finnish.

In addition to Vake (currently the Climate Fund) and Solita, the Finnish National Broadcasting Company Yle and the University of Helsinki participated in the project. Solita was responsible for project planning and implementation.

Bringing Finns together to help develop speech recognition

The initial objective was to develop Finnish speech recognition and high-quality digital services openly and securely, with Finnish resources, and to encourage Finnish people to participate in building services that can understand Finnish better.

The idea was to make donating speech samples easy, absolutely secure and confidential, but also interesting and entertaining, to collect a sufficient volume of samples. The implementation was also technically accessible: Speech samples can be donated using most phones and internet browsers. Meeting high quality requirements, the system also takes into account users with special needs. For example, visually impaired users can also donate samples.

The project kick-off was preceded by a six-month planning period, bringing together experts from Vake, Yle, and Solita, linguists from the University of Helsinki, and various other partner businesses. During the planning period, a concept and an implementation plan were devised.

Donations

  • 4000 Speech samples donated: more than 4,000 hours
  • 214,000 Individual donations: 214,000

Groups presented

  • Different dialects

  • Age groups

  • Genders

  • Immigrants

  • Visually and speech-impaired

The project kick-off was preceded by a six-month planning period, bringing together experts from Vake, Yle, and Solita, linguists from the University of Helsinki and various other partner businesses. During the planning period, a concept and an implementation plan were devised.

Would you like to give a sports commentary on the long-distance runner Lasse Virén?

The goal of the Donate speech project is to collect 10,000 hours of free-form everyday speech in Finnish. For a machine to learn to understand Finnish, a large pool of speech samples comprising a wide range of spoken language varieties is required: the speech bank should contain different dialects, samples from different genders and age groups, and feature groups such as immigrants. The Donate speech web service where you can easily donate your own speech samples was built to help achieve this goal.

How can speech samples be collected easily, inspiringly, and efficiently? To help users produce speech, Solita and Yle worked together to build various prompts into the system. Users can donate speech using either a mobile application or the web service. Within the service, you can select various themes and interview questions: your sample can be anything from telling about you day to giving a sports commentary on Lasse Virén. Between collecting samples, the service will also ask users for background information such as their age.

The visibility of the service was boosted with Yle TV campaigns, media communications, and researcher collaboration.

High-standard implementation as a quality export product

The collection of speech samples commenced in June 2020. Today, anyone can donate their own Finnish speech samples at lahjoitapuhetta.fi or by downloading the Donate Speech mobile app. Individual users cannot be identified based on their speech samples, and the content of the samples is irrelevant for the project. Users are encouraged to keep it casual – no confidential information should be included in the recordings.

Download the Donate speech app from the App Store or Google Play Store

In terms of its technical implementation, the project was characterised by an exceptionally high standard regarding quality and data security. In terms of user experience, the goal was to make donating, i.e. talking to a recorder, as fun and inspiring as possible for a variety of Finnish speakers.

The donated speech samples are stored in the University of Helsinki speech bank, and from there, they can be distributed to researchers and companies working on applications employing speech recognition.

The concept was designed to be modular and flexible. The implementation of the project can easily be adapted to suit changing needs, and in the future, the concept could potentially be productised for export.

The feedback from linguists has been overwhelmingly positive so far. The visual design implemented by Yle and Solita has also been well received as visually pleasing and clear.

The sky is the limit – kudos from researchers and businesses

Donate speech is a socially significant project that helps researchers and businesses develop high-impact new Finnish-language services for healthcare, education, and the care of the elderly. Only the sky is the limit for building services that are socially significant and improve the quality of everyday life using the databank.

Over 4,000 hours of speech have been collected so far, and more is being donated every day. So far, there have been over 214,000 individual donations, and the elderly in Finland in particular have been very active donors.

For the successful development of machine learning and AI applications, it is also important to collect various metadata, such as the speech donors’ place of birth, dialect, age, and gender. The service only collects data provided by the users, and no other usage data is stored.

In the future, the constantly growing and improving speech bank will help improve the everyday quality of life of every Finn. Enhancing Finnish-language speech recognition will not just benefit professionals such as linguists and doctors, but everyone in Finland.

Customer: Finnish State Development Company Vake (currently the Climate Fund)
Design: Solita, Vake, Univesity of Helsinki, 1001 Lakes, Yle, Aalto University, University of Turku, Onerva Hoiva
ImplementationSolita, Yle

Grand One competition 2021

  1. Best Mobile Service

  2. Best Use of Data, Honorable Mention