Robustness of Self-Supervised Speech Models on Flemish Dutch vs. Multilingual Models in Noisy and Accented Conditions

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 robustness of self-supervised speech models pre-trained on Flemish Dutch compare to multilingual models when tested on noisy or accented speech conditions, as measured by word error rate on the CHiME or AMI datasets?

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