Inference Efficiency Trade-offs of Flemish Dutch Versus High-Resource Germanic Self-Supervised Speech Models on Edge Devices

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 inference efficiency trade-off (measured in latency and throughput) between self-supervised speech models pre-trained on Flemish Dutch and those pre-trained on high-resource Germanic languages when deployed in edge devices for on-device speech recognition tasks?

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