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 Tools annual

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 Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Automatic Speech Recognition (ASR) systems frequently use a search-baseddecoding strategy aiming to find the best attainable transcript by consideringmultiple candidates. One prominent speech recognition decoding heuristic isbeam search, which seeks the transcript with the greatest likelihood computedusing the predicted distribution. While showing substantial performance gainsin various tasks, beam search loses some of its effectiveness when thepredicted probabilities are highly confident, i.e., the predicted distributionis massed for a single or very few classes. We show that recently proposedSelf-Supervised Learning (SSL)-based ASR models tend to yield exceptionallyconfident predictions that may hamper beam search from truly considering adiverse set of candidates. We perform a layer analysis to reveal and visualizehow predictions evolve, and propose a decoding procedure that improves theperformance of fine-tuned ASR models. Our proposed approach does not requirefurther training beyond the original fine-tuning, nor additional modelparameters. In fact, we find that our proposed method requires significantlyless inference computation than current approaches. We propose aggregating thetop M layers, potentially leveraging useful information encoded in intermediatelayers, and relaxing model confidence. We demonstrate the effectiveness of ourapproach by conducting an empirical study on varying amounts of labeledresources and different model sizes, showing consistent improvements inparticular when applied to low-resource scenarios.

Don't Be So Sure! Boosting ASR Decoding via Confidence Relaxation