Fine-tuning Performance of English-only ASR Models on Flemish Dutch versus Low-resource Germanic Languages

Recent research in speech processing exhibits a growing interest in unsupervised and self-supervised representation learning from unlabelled data to alleviate the need for large amounts of annotated data. We investigate several popular pre-training methods and apply them to Flemish Dutch. We compare off-the-shelf English pre-trained models to models trained on an increasing amount of Flemish data. We find that the most important factors for positive transfer to downstream speech recognition tasks include a substantial amount of data and a matching pre-training domain. Ideally, we also finetune

Research goal: How does the fine-tuning performance of English-only pre-trained ASR models on Flemish Dutch compare to models pre-trained on other low-resource Germanic languages like Afrikaans or Luxembourgish, measured by WER on CommonVoice datasets?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.5/10.