Domain-Adapted Fine-Tuning of Flemish Dutch Speech Models for CommonVoice Benchmark 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 fine-tuning Flemish Dutch self-supervised speech models with domain adaptation techniques affect word error rate on the CommonVoice benchmark compared to direct cross-lingual transfer from English pre-trained models?

Autonomous synthesis report generated by Assignee Research. Tribunal consensus score: 7.7/10.