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 Tools annual

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 monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

This work studies knowledge distillation (KD) and addresses its constraintsfor recurrent neural network transducer (RNN-T) models. In hard distillation, ateacher model transcribes large amounts of unlabelled speech to train a studentmodel. Soft distillation is another popular KD method that distills the outputlogits of the teacher model. Due to the nature of RNN-T alignments, applyingsoft distillation between RNN-T architectures having different posteriordistributions is challenging. In addition, bad teachers having highword-error-rate (WER) reduce the efficacy of KD. We investigate how toeffectively distill knowledge from variable quality ASR teachers, which has notbeen studied before to the best of our knowledge. We show that a sequence-levelKD, full-sum distillation, outperforms other distillation methods for RNN-Tmodels, especially for bad teachers. We also propose a variant of full-sumdistillation that distills the sequence discriminative knowledge of the teacherleading to further improvement in WER. We conduct experiments on publicdatasets namely SpeechStew and LibriSpeech, and on in-house production data.

Robust Knowledge Distillation from RNN-T Models With Noisy Training Labels Using Full-Sum Loss