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

While the community keeps promoting end-to-end models over conventionalhybrid models, which usually are long short-term memory (LSTM) models trainedwith a cross entropy criterion followed by a sequence discriminative trainingcriterion, we argue that such conventional hybrid models can still besignificantly improved. In this paper, we detail our recent efforts to improveconventional hybrid LSTM acoustic models for high-accuracy and low-latencyautomatic speech recognition. To achieve high accuracy, we use a contextuallayer trajectory LSTM (cltLSTM), which decouples the temporal modeling andtarget classification tasks, and incorporates future context frames to get moreinformation for accurate acoustic modeling. We further improve the trainingstrategy with sequence-level teacher-student learning. To obtain low latency,we design a two-head cltLSTM, in which one head has zero latency and the otherhead has a small latency, compared to an LSTM. When trained with Microsoft's 65thousand hours of anonymized training data and evaluated with test sets with1.8 million words, the proposed two-head cltLSTM model with the proposedtraining strategy yields a 28.2\% relative WER reduction over the conventionalLSTM acoustic model, with a similar perceived latency.

High-Accuracy and Low-Latency Speech Recognition with Two-Head Contextual Layer Trajectory LSTM Model