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

Environmental Sound Classification (ESC) is a challenging field of researchin non-speech audio processing. Most of current research in ESC focuses ondesigning deep models with special architectures tailored for specific audiodatasets, which usually cannot exploit the intrinsic patterns in the data.However recent studies have surprisingly shown that transfer learning frommodels trained on ImageNet is a very effective technique in ESC. Herein, wepropose SoundCLR, a supervised contrastive learning method for effectiveenvironment sound classification with state-of-the-art performance, which worksby learning representations that disentangle the samples of each class fromthose of other classes. Our deep network models are trained by combining acontrastive loss that contributes to a better probability output by theclassification layer with a cross-entropy loss on the output of the classifierlayer to map the samples to their respective 1-hot encoded labels. Due to thecomparatively small sizes of the available environmental sound datasets, wepropose and exploit a transfer learning and strong data augmentation pipelineand apply the augmentations on both the sound signals and their log-melspectrograms before inputting them to the model. Our experiments show that ourmasking based augmentation technique on the log-mel spectrograms cansignificantly improve the recognition performance. Our extensive benchmarkexperiments show that our hybrid deep network models trained with combinedcontrastive and cross-entropy loss achieved the state-of-the-art performance onthree benchmark datasets ESC-10, ESC-50, and US8K with validation accuracies of99.75\%, 93.4\%, and 86.49\% respectively. The ensemble version of our modelsalso outperforms other top ensemble methods. The code is available athttps://github.com/alireza-nasiri/SoundCLR.

SoundCLR: Contrastive Learning of Representations For Improved Environmental Sound Classification