There’s been a steady progress in the accuracy and performance of automatic speech recognition and synthesis but challenges remain as to capturing the rich, complex human spoken language. In this project, we propose bonding academic and industrial partners to address the issue of the lack of developments in the area of automatic speech recognition of the spoken dialects of Swedish in Finnish territory. Our goal is to gather open-access labelled speech dialect data for the Swedish speaking population from across Finland to develop a set of ASR technologies and then test them in the field. The project aims at addressing this general, as well as regional, gap in speech recognition as we will advance speech recognition in the Swedish-Finnish domain. We adopt a human-centered co-creation approach, where we collect speech data as well as test the developed speech algorithm out in the field. Persons, whose mother tongue is the tested dialect, evaluate how they experience the speech synthesis/recognition in a healthcare context. The gathering and labelling of speech data will be done for six different Finnish Swedish dialects: 1. Åland 2. Pargas 3. Södra Helsingfors 4. Närpes 5. Korsholm (e.g, Kvevlax, Replot) 6. Borgå Deliverables - Open source Swedish data-set for researchers and companies - Pre-trained speech recognition model for Swedish spoken in Finland - Testing algorithm in real use environment - Research paper

2020