Keyword spotting for audiovisual archival search in Uralic languages

Abstract: In this study we investigate the potential of using Automatic Speech Recognition (ASR) for keyword spotting for four Uralic languages: Finnish, Hungarian, Estonian and Komi. These languages also represent different levels on the high and low resource continuum. Although the accuracy of the ASR systems show there is a long way to go, we show that they still have potential to be useful for downstream tasks such as key-word spotting. By using a simple text search after running ASR, we are already able to achieve an F 1 score of between 0.15 and 0.33, a precision of nearly 0.90 for Estonian and Hungarian, and a precision of 0.76 for Komi.