Varied English-Flemish Data Ratios in Mixed Pre-Training and Low-Resource ASR Performance

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: What is the impact of varying the ratio of English-to-Flemish data in mixed pre-training datasets on the WER performance of downstream low-resource European language speech recognition tasks?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.0/10.