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

The multi-decoder (MD) end-to-end speech translation model has demonstratedhigh translation quality by searching for better intermediate automatic speechrecognition (ASR) decoder states as hidden intermediates (HI). It is a two-passdecoding model decomposing the overall task into ASR and machine translationsub-tasks. However, the decoding speed is not fast enough for real-worldapplications because it conducts beam search for both sub-tasks duringinference. We propose Fast-MD, a fast MD model that generates HI bynon-autoregressive (NAR) decoding based on connectionist temporalclassification (CTC) outputs followed by an ASR decoder. We investigated twotypes of NAR HI: (1) parallel HI by using an autoregressive Transformer ASRdecoder and (2) masked HI by using Mask-CTC, which combines CTC and theconditional masked language model. To reduce a mismatch in the ASR decoderbetween teacher-forcing during training and conditioning on CTC outputs duringtesting, we also propose sampling CTC outputs during training. Experimentalevaluations on three corpora show that Fast-MD achieved about 2x and 4x fasterdecoding speed than that of the na\"ive MD model on GPU and CPU with comparabletranslation quality. Adopting the Conformer encoder and intermediate CTC lossfurther boosts its quality without sacrificing decoding speed.

Fast-MD: Fast Multi-Decoder End-to-End Speech Translation with Non-Autoregressive Hidden Intermediates