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 Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

In this paper, we present MixRep, a simple and effective data augmentationstrategy based on mixup for low-resource ASR. MixRep interpolates the featuredimensions of hidden representations in the neural network that can be appliedto both the acoustic feature input and the output of each layer, whichgeneralizes the previous MixSpeech method. Further, we propose to combine themixup with a regularization along the time axis of the input, which is shown ascomplementary. We apply MixRep to a Conformer encoder of an E2E LASarchitecture trained with a joint CTC loss. We experiment on the WSJ datasetand subsets of the SWB dataset, covering reading and telephony conversationalspeech. Experimental results show that MixRep consistently outperforms otherregularization methods for low-resource ASR. Compared to a strong SpecAugmentbaseline, MixRep achieves a +6.5\% and a +6.7\% relative WER reduction on theeval92 set and the Callhome part of the eval'2000 set.

MixRep: Hidden Representation Mixup for Low-Resource Speech Recognition