This project was carried out between 2015 and 2018. In the project we disclosed over 2500 hours of radio broadcasts from the Omrop Fryslân (Frisian Broadcast). The radio broadcasts contain spoken Frisian and Dutch covering the period 1950–2000. We are working on a follow-up.
We used speech technology for spoken document retrieval (speech to text conversion) and for speaker tracking (speaker diarization & recognition). Thus we were able to locate broadcasts addressing specific topics and specific speakers in the audio signal. In order to guarantee relevance in retrieval, the project also developed an enriched Frisian lexicon and a semantic search engine for Frisian and Dutch to search the broadcasts. The non-academic project partners acknowledge the disclosure of this data as a rich source of Frisian cultural heritage. The project carried out innovative research since it investigated efficiency and performance of: 1. Automatic Speech Recognition of Frisian and Dutch using either two separate recognizers or a hybrid one; 2. the integration of speaker diarization and speaker recognition applied to a large longitudinal data set; 3. a flexible semantic search interface targeted at various user groups. In all these topics efficient processing is required, because of the sheer volume of the data.
FAME! stands for Frisian Audio Mining Enterprise. The project is funded by NWO’s Creative Industry Programme under project number 314-99-119. See here.