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

While much of modern speech and audio processing relies on deep neuralnetworks trained using fixed audio representations, recent studies suggestgreat potential in acoustic frontends learnt jointly with a backend. In thisstudy, we focus specifically on learnable filterbanks. Prior studies havereported that in frontends using learnable filterbanks initialised to a melscale, the learned filters do not differ substantially from theirinitialisation. Using a Gabor-based filterbank, we investigate the sensitivityof a learnable filterbank to its initialisation using several initialisationstrategies on two audio tasks: voice activity detection and bird speciesidentification. We use the Jensen-Shannon Distance and analysis of the learnedfilters before and after training. We show that although performance is overallimproved, the filterbanks exhibit strong sensitivity to their initialisationstrategy. The limited movement from initialised values suggests that alternateoptimisation strategies may allow a learnable frontend to reach better overallperformance.

Learnable Frontends that do not Learn: Quantifying Sensitivity to Filterbank Initialisation