Developing Live Subtitling in Finland

Moving from manual subtitling towards respeaking

Authors

• Kaisa Vitikainen University of Helsinki

DOI:

https://doi.org/10.61200/mikael.129389

Keywords:

Live subtitling, subtitling for the deaf and hard of hearing, SDH, respeaking, speech recognition

Abstract

Respeaking with speech recognition software is the preferred way of producing live subtitles worldwide, but it is not used at all in Finland. All live subtitling (the intralingual subtitling of live programs) in Finland is done manually. This paper presents a pilot study on respeaking in Finnish, comparing authentic manual subtitles to simulated respoken subtitles and assessing their accuracy through the NER model. While the accuracy rate of respeaking does not reach the ideal 98%, it is found to be higher than that of manual subtitles, which suffer from significant loss of content. With trained respeakers and some improvements to the speech recognition software, respeaking in Finnish could reach 98% accuracy, and is worth pursuing further.