A Speech Recognizer for Frisian/Dutch Council Meetings

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 1009–1015 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 1009 A Speech Recognizer for Frisian/Dutch Council Meetings Martijn Bentum1, Louis ten Bosch1, Henk van den Heuvel1, Simone Wills1, Domenique van der Niet2, Jelske Dijkstra3, Hans Van de Velde3,4 1CLS/CLST, Radboud University; 2Humain’r; 3Fryske Akademy; 4Utrecht University {martijn.bentum, louis.tenbosch, henk.vandenheuvel, simone.wills}@ru.nl, domenique@humainr.com, {jdijkstra,hvandevelde}@fryske-akademy.nl Abstract We developed a bilingual Frisian/Dutch speech recognizer for council meetings in Fryslân (the Netherlands). During these meetings both Frisian and Dutch are spoken, and code switching between both languages shows up frequently. The new speech recognizer is based on an existing speech recognizer for Frisian and Dutch named FAME!, which was trained and tested on historical radio broadcasts. Adapting a speech recognizer for the council meeting domain is challenging because of acoustic background noise, speaker overlap and the jargon typically used in council meetings. To train the new recognizer, we used the radio broadcast materials utilized for the development of the FAME! recognizer and added newly created manually transcribed audio recordings of council meetings from eleven Frisian municipalities, the Frisian provincial council and the Frisian water board. The council meeting recordings consist of 49 hours of speech, with 26 hours of Frisian speech and 23 hours of Dutch speech. Furthermore, from the same sources, we obtained texts in the domain of council meetings containing 11 million words; 1.1 million Frisian words and 9.9 million Dutch words. We describe the methods used to train the new recognizer, report the observed word error rates, and perform an error analysis on remaining errors. Keywords: ASR, Code Switching, Frisian, Dutch, Domain Adaptation 1. Introduction We developed a new bilingual Frisian/Dutch automatic speech recognizer (ASR) for Frisian council meetings. During these council meetings both the Frisian and the Dutch language are spoken. We build on an existing Frisian/Dutch ASR system, entitled FAME! (Yılmaz et al., 2018), which was trained and tested on radio broadcasts. This FAME! speech recognizer was adapted for the domain of council meetings. Adapting the ASR system for the Frisian council meeting domain is challenging because of the noisy background, the use of jargon and multiple overlapping speakers. During these meetings, long and complex words are used, typical for civil servants and government style documents, but not part of a general-purpose lexicon. For example, ambtenaarûndersteuning ‘civil servant support’ is a compound consisting of the stems ambtenaar ‘civil servant’ and ûndersteun ‘support’, or kostprijsberekkening ‘cost price calculation’ is a compound consisting of the stems kost ‘cost’, prijs ‘price’ and berekken ‘calculate’. For this project an automatic speech recognizer was built that could deal with Frisian and Dutch as well as with code switching between the two languages as is often encountered in Frisian. We trained a new speech recognizer based on the FAME! radio broadcast materials combined with new speech recordings. The FAME! materials (https://fame.ruhosting.nl/) contain more than 3,000 hours of Frisian and Dutch radio broadcasts from the Omrop Fryslân (Frisian Broadcast) …

Content truncated to 500 words