Large scale weakly and semi-supervised learning for low-resource video ASR
This work addresses labeling costs for speech recognition in low-resource languages, but it is incremental as it systematically compares existing methods.
The paper tackled the problem of transcribing low-resource social media videos by comparing self-labeling and weakly-supervised pretraining methods, resulting in a 20% relative WER reduction for encoder-decoder models compared to supervised baselines.
Many semi- and weakly-supervised approaches have been investigated for overcoming the labeling cost of building high quality speech recognition systems. On the challenging task of transcribing social media videos in low-resource conditions, we conduct a large scale systematic comparison between two self-labeling methods on one hand, and weakly-supervised pretraining using contextual metadata on the other. We investigate distillation methods at the frame level and the sequence level for hybrid, encoder-only CTC-based, and encoder-decoder speech recognition systems on Dutch and Romanian languages using 27,000 and 58,000 hours of unlabeled audio respectively. Although all approaches improved upon their respective baseline WERs by more than 8%, sequence-level distillation for encoder-decoder models provided the largest relative WER reduction of 20% compared to the strongest data-augmented supervised baseline.