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

Connectionist Temporal Classification (CTC) is a widely used criterion fortraining supervised sequence-to-sequence (seq2seq) models. It enables learningthe relations between input and output sequences, termed alignments, bymarginalizing over perfect alignments (that yield the ground truth), at theexpense of imperfect alignments. This binary differentiation of perfect andimperfect alignments falls short of capturing other essential alignmentproperties that hold significance in other real-world applications. Here wepropose $\textit{Align With Purpose}$, a $\textbf{general Plug-and-Playframework}$ for enhancing a desired property in models trained with the CTCcriterion. We do that by complementing the CTC with an additional loss termthat prioritizes alignments according to a desired property. Our method doesnot require any intervention in the CTC loss function, enables easyoptimization of a variety of properties, and allows differentiation betweenboth perfect and imperfect alignments. We apply our framework in the domain ofAutomatic Speech Recognition (ASR) and show its generality in terms of propertyselection, architectural choice, and scale of training dataset (up to 280,000hours). To demonstrate the effectiveness of our framework, we apply it to twounrelated properties: emission time and word error rate (WER). For the former,we report an improvement of up to 570ms in latency optimization with a minorreduction in WER, and for the latter, we report a relative improvement of 4.5%WER over the baseline models. To the best of our knowledge, these applicationshave never been demonstrated to work on a scale of data as large as ours.Notably, our method can be implemented using only a few lines of code, and canbe extended to other alignment-free loss functions and to domains other thanASR.

Align With Purpose: Optimize Desired Properties in CTC Models with a General Plug-and-Play Framework