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

End-to-end models have achieved state-of-the-art results on several automaticspeech recognition tasks. However, they perform poorly when evaluated onlong-form data, e.g., minutes long conversational telephony audio. One reasonthe model fails on long-form speech is that it has only seen short utterancesduring training. In this paper we study the effect of training utterance lengthon the word error rate (WER) for RNN-transducer (RNN-T) model. We compare twowidely used training objectives, log loss (or RNN-T loss) and minimum worderror rate (MWER) loss. We conduct experiments on telephony datasets in fourlanguages. Our experiments show that for both losses, the WER on long-formspeech reduces substantially as the training utterance length increases. Theaverage relative WER gain is 15.7% for log loss and 8.8% for MWER loss. Whentraining on short utterances, MWER loss leads to a lower WER than the log loss.Such difference between the two losses diminishes when the input lengthincreases.

Input Length Matters: Improving RNN-T and MWER Training for Long-form Telephony Speech Recognition