Evaluating Standard and Dialectal Frisian ASR: Multilingual Fine-tuning and Language Identification for Improved Low-resource Performance
Reihaneh Amooie, Wietse de Vries, Yun Hao, Jelske Dijkstra, Matt Coler, Martijn Wieling
TL;DR
This paper tackles ASR for Frisian, a low-resource language with notable dialectal variation, by exploring multilingual self-supervised pretraining and fine-tuning augmented with language identification. Using the XLS-R 1B model, the authors fine-tune Frisian data in combination with Dutch, German, and English, and inject an LID token to condition transcription. They find that Dutch and German aid Frisian ASR on standard data, while English can hurt performance; LID yields modest gains on standard data but clearer benefits for dialectal SPRAAKLAB data, and dialect performance is strongly influenced by how dialects are elicited. The study suggests that evaluating only standard Frisian data underestimates real-world performance in dialect-rich settings and underscores the value of multilingual, context-aware approaches for improving ASR in minority languages.
Abstract
Automatic Speech Recognition (ASR) performance for low-resource languages is still far behind that of higher-resource languages such as English, due to a lack of sufficient labeled data. State-of-the-art methods deploy self-supervised transfer learning where a model pre-trained on large amounts of data is fine-tuned using little labeled data in a target low-resource language. In this paper, we present and examine a method for fine-tuning an SSL-based model in order to improve the performance for Frisian and its regional dialects (Clay Frisian, Wood Frisian, and South Frisian). We show that Frisian ASR performance can be improved by using multilingual (Frisian, Dutch, English and German) fine-tuning data and an auxiliary language identification task. In addition, our findings show that performance on dialectal speech suffers substantially, and, importantly, that this effect is moderated by the elicitation approach used to collect the dialectal data. Our findings also particularly suggest that relying solely on standard language data for ASR evaluation may underestimate real-world performance, particularly in languages with substantial dialectal variation.