English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

While Automatic Speech Recognition (ASR) models have shown significantadvances with the introduction of unsupervised or self-supervised trainingtechniques, these improvements are still only limited to a subsection oflanguages and speakers. Transfer learning enables the adaptation of large-scalemultilingual models to not only low-resource languages but also to morespecific speaker groups. However, fine-tuning on data from new domains isusually accompanied by a decrease in performance on the original domain.Therefore, in our experiments, we examine how well the performance oflarge-scale ASR models can be approximated for smaller domains, with our owndataset of German Senior Voice Commands (SVC-de), and how much of the generalspeech recognition performance can be preserved by selectively freezing partsof the model during training. To further increase the robustness of the ASRmodel to vocabulary and speakers outside of the fine-tuned domain, we applyExperience Replay for continual learning. By adding only a fraction of datafrom the original domain, we are able to reach Word-Error-Rates (WERs) below5\% on the new domain, while stabilizing performance for general speechrecognition at acceptable WERs.

Replay to Remember: Continual Layer-Specific Fine-tuning for German Speech Recognition