Performance comparison of self-supervised speech models on Flemish Dutch and English data for LibriSpeech WER in noise

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 performance of self-supervised speech models trained on Flemish Dutch data compare to those trained on English data when evaluated on the LibriSpeech benchmark for word error rate (WER) in noisy environments?

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