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

Non-native speech causes automatic speech recognition systems to degrade inperformance. Past strategies to address this challenge have considered modeladaptation, accent classification with a model selection, alternatepronunciation lexicon, etc. In this study, we consider a recurrent neuralnetwork (RNN) with connectionist temporal classification (CTC) cost functiontrained on multi-accent English data including US (Native), Indian and Hispanicaccents. We exploit dark knowledge from a model trained with the multi-accentdata to train student models under the guidance of both a teacher model and CTCcost of target transcription. We show that transferring knowledge from a singleRNN-CTC trained model toward a student model, yields better performance thanthe stand-alone teacher model. Since the outputs of different trained CTCmodels are not necessarily aligned, it is not possible to simply use anensemble of CTC teacher models. To address this problem, we train accentspecific models under the guidance of a single multi-accent teacher, whichresults in having multiple aligned and trained CTC models. Furthermore, wetrain a student model under the supervision of the accent-specific teachers,resulting in an even further complementary model, which achieves +20.1%relative Character Error Rate (CER) reduction compared to the baseline trainedwithout any teacher. Having this effective multi-accent model, we can achievefurther improvement for each accent by adapting the model to each accent. Usingthe accent specific model's outputs to regularize the adapting process (i.e., aknowledge distillation version of Kullback-Leibler (KL) divergence) results ineven superior performance compared to the conventional approach using generalteacher models.

Advancing Multi-Accented LSTM-CTC Speech Recognition using a Domain Specific Student-Teacher Learning Paradigm