🤖 AI Summary
Swedish, as a medium-resource language, suffers from suboptimal performance in multilingual ASR models (e.g., Whisper) due to insufficient representation in training data. To address this, we construct the first high-diversity, cross-domain monolingual Swedish training set—built upon the largest publicly available Swedish speech corpus to date—and systematically perform supervised fine-tuning of the Whisper architecture. Our methodology incorporates rigorous data cleaning and augmentation strategies to significantly enhance model robustness. Evaluation across three major benchmarks—FLEURS, CoVo, and NST—demonstrates that our fine-tuned model achieves an average 47% relative WER reduction over Whisper-large-v3, consistently outperforming the baseline across all model sizes. This work breaks the fine-tuning performance ceiling for medium-resource languages in ASR and establishes a reproducible, high-quality monolingual data-driven paradigm for model optimization.
📝 Abstract
This work presents a suite of fine-tuned Whisper models for Swedish, trained on a dataset of unprecedented size and variability for this mid-resourced language. As languages of smaller sizes are often underrepresented in multilingual training datasets, substantial improvements in performance can be achieved by fine-tuning existing multilingual models, as shown in this work. This work reports an overall improvement across model sizes compared to OpenAI's Whisper evaluated on Swedish. Most notably, we report an average 47% reduction in WER comparing our best performing model to OpenAI's whisper-large-v3, in evaluations across FLEURS, Common Voice, and NST.