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

Conventional time-delay neural networks (TDNNs) struggle to handle long-rangecontext, their ability to represent speaker information is therefore limited inlong utterances. Existing solutions either depend on increasing modelcomplexity or try to balance between local features and global context toaddress this issue. To effectively leverage the long-term dependencies of audiosignals and constrain model complexity, we introduce a novel module calledGlobal-aware Filter layer (GF layer) in this work, which employs a set oflearnable transform-domain filters between a 1D discrete Fourier transform andits inverse transform to capture global context. Additionally, we develop adynamic filtering strategy and a sparse regularization method to enhance theperformance of the GF layer and prevent overfitting. Based on the GF layer, wepresent a dual-stream TDNN architecture called DS-TDNN for automatic speakerverification (ASV), which utilizes two unique branches to extract both localand global features in parallel and employs an efficient strategy to fusedifferent-scale information. Experiments on the Voxceleb and SITW databasesdemonstrate that the DS-TDNN achieves a relative improvement of 10\% togetherwith a relative decline of 20\% in computational cost over the ECAPA-TDNN inspeaker verification task. This improvement will become more evident as theutterance's duration grows. Furthermore, the DS-TDNN also beats popular deepresidual models and attention-based systems on utterances of arbitrary length.

DS-TDNN: Dual-stream Time-delay Neural Network with Global-aware Filter for Speaker Verification