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

We propose a multi-channel speech enhancement approach with a novel two-stagefeature fusion method and a pre-trained acoustic model in a multi-task learningparadigm. In the first fusion stage, the time-domain and frequency-domainfeatures are extracted separately. In the time domain, the multi-channelconvolution sum (MCS) and the inter-channel convolution differences (ICDs)features are computed and then integrated with the first 2-D convolutionallayer, while in the frequency domain, the log-power spectra (LPS) features fromboth original channels and super-directive beamforming outputs are combinedwith a second 2-D convolutional layer. To fully integrate the rich informationof multi-channel speech, i.e. time-frequency domain features and the arraygeometry, we apply a third 2-D convolutional layer in the second fusion stageto obtain the final convolutional features. Furthermore, we propose to use afixed clean acoustic model trained with the end-to-end lattice-free maximummutual information criterion to enforce the enhanced output to have the samedistribution as the clean waveform to alleviate the over-estimation problem ofthe enhancement task and constrain distortion. On the Task1 development datasetof ConferencingSpeech 2021 challenge, a PESQ improvement of 0.24 and 0.19 isattained compared to the official baseline and a recently proposedmulti-channel separation method.

Multi-channel Speech Enhancement with 2-D Convolutional Time-frequency Domain Features and a Pre-trained Acoustic Model