Scaling Self-Supervised Speech Models for Flemish Dutch Downstream 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: How does the scaling of self-supervised speech model size affect downstream task performance (e.g., WER) when pre-trained on Flemish Dutch data, and what is the optimal model size for resource-constrained settings?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 8.5/10.