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

RNN-T is currently considered the industry standard in ASR due to itsexceptional WERs in various benchmark tests and its ability to support seamlessstreaming and longform transcription. However, its biggest drawback lies in thesignificant discrepancy between its training and inference objectives. Duringtraining, RNN-T maximizes all alignment probabilities by teacher forcing, whileduring inference, it uses beam search which may not necessarily find themaximum probable alignment. Additionally, RNN-T's inability to experiencemistakes during teacher forcing training makes it more problematic when amistake occurs in inference. To address this issue, this paper proposes aReinforcement Learning method that minimizes the gap between training andinference time. Our Edit Distance based RL (EDRL) approach computes rewardsbased on the edit distance, and trains the network at every action level. Theproposed approach yielded SoTA WERs on LibriSpeech for the 600M Conformer RNN-Tmodel.

Edit Distance based RL for RNNT decoding