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

Training deep neural networks on well-understood dependencies in speech datacan provide new insights into how they learn internal representations. Thispaper argues that acquisition of speech can be modeled as a dependency betweenrandom space and generated speech data in the Generative Adversarial Networkarchitecture and proposes a methodology to uncover the network's internalrepresentations that correspond to phonetic and phonological properties. TheGenerative Adversarial architecture is uniquely appropriate for modelingphonetic and phonological learning because the network is trained onunannotated raw acoustic data and learning is unsupervised without anylanguage-specific assumptions or pre-assumed levels of abstraction. AGenerative Adversarial Network was trained on an allophonic distribution inEnglish. The network successfully learns the allophonic alternation: thenetwork's generated speech signal contains the conditional distribution ofaspiration duration. The paper proposes a technique for establishing thenetwork's internal representations that identifies latent variables thatcorrespond to, for example, presence of [s] and its spectral properties. Bymanipulating these variables, we actively control the presence of [s] and itsfrication amplitude in the generated outputs. This suggests that the networklearns to use latent variables as an approximation of phonetic and phonologicalrepresentations. Crucially, we observe that the dependencies learned intraining extend beyond the training interval, which allows for additionalexploration of learning representations. The paper also discusses how thenetwork's architecture and innovative outputs resemble and differ fromlinguistic behavior in language acquisition, speech disorders, and speecherrors, and how well-understood dependencies in speech data can help usinterpret how neural networks learn their representations.

Generative Adversarial Phonology: Modeling unsupervised phonetic and phonological learning with neural networks