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 Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

This paper describes the IDLab submission for the text-independent task ofthe Short-duration Speaker Verification Challenge 2021 (SdSVC-21). This speakerverification competition focuses on short duration test recordings andcross-lingual trials, along with the constraint of limited availability ofin-domain DeepMine Farsi training data. Currently, both Time Delay NeuralNetworks (TDNNs) and ResNets achieve state-of-the-art results in speakerverification. These architectures are structurally very different and theconstruction of hybrid networks looks a promising way forward. We introduce a2D convolutional stem in a strong ECAPA-TDNN baseline to transfer some of thestrong characteristics of a ResNet based model to this hybrid CNN-TDNNarchitecture. Similarly, we incorporate absolute frequency positional encodingsin an SE-ResNet34 architecture. These learnable feature map biases along thefrequency axis offer this architecture a straightforward way to exploitfrequency positional information. We also propose a frequency-wise variant ofSqueeze-Excitation (SE) which better preserves frequency-specific informationwhen rescaling the feature maps. Both modified architectures significantlyoutperform their corresponding baseline on the SdSVC-21 evaluation data and theoriginal VoxCeleb1 test set. A four system fusion containing the two improvedarchitectures achieved a third place in the final SdSVC-21 Task 2 ranking.

Integrating Frequency Translational Invariance in TDNNs and Frequency Positional Information in 2D ResNets to Enhance Speaker Verification