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

In this paper, we describe the Maximum Uniformity of Distribution (MUD)algorithm with the power-law nonlinearity. In this approach, we hypothesizethat neural network training will become more stable if feature distribution isnot too much skewed. We propose two different types of MUD approaches: powerfunction-based MUD and histogram-based MUD. In these approaches, we firstobtain the mel filterbank coefficients and apply nonlinearity functions foreach filterbank channel. With the power function-based MUD, we apply apower-function based nonlinearity where power function coefficients are chosento maximize the likelihood assuming that nonlinearity outputs follow theuniform distribution. With the histogram-based MUD, the empirical CumulativeDensity Function (CDF) from the training database is employed to transform theoriginal distribution into a uniform distribution. In MUD processing, we do notuse any prior knowledge (e.g. logarithmic relation) about the energy of theincoming signal and the perceived intensity by a human. Experimental resultsusing an end-to-end speech recognition system demonstrate that power-functionbased MUD shows better result than the conventional Mel Filterbank CepstralCoefficients (MFCCs). On the LibriSpeech database, we could achieve 4.02 % WERon test-clean and 13.34 % WER on test-other without using any Language Models(LMs). The major contribution of this work is that we developed a new algorithmfor designing the compressive nonlinearity in a data-driven way, which is muchmore flexible than the previous approaches and may be extended to other domainsas well.

power-law nonlinearity with maximally uniform distribution criterion for improved neural network training in automatic speech recognition