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

Target sound detection (TSD) aims to detect the target sound from mixtureaudio given the reference information. Previous works have shown that TSDmodels can be trained on fully-annotated (frame-level label) orweakly-annotated (clip-level label) data. However, there are some clearevidences show that the performance of the model trained on weakly-annotateddata is worse than that trained on fully-annotated data. To fill this gap, weprovide a mixed supervision perspective, in which learning novel categories(target domain) using weak annotations with the help of full annotations ofexisting base categories (source domain). To realize this, a mixed supervisedlearning framework is proposed, which contains two mutually-helping studentmodels (\textit{f\_student} and \textit{w\_student}) that learn fromfully-annotated and weakly-annotated data, respectively. The motivation is that\textit{f\_student} learned from fully-annotated data has a better ability tocapture detailed information than \textit{w\_student}. Thus, we first let\textit{f\_student} guide \textit{w\_student} to learn the ability to capturedetails, so \textit{w\_student} can perform better in the target domain. Thenwe let \textit{w\_student} guide \textit{f\_student} to fine-tune on the targetdomain. The process can be repeated several times so that the two studentsperform very well in the target domain. To evaluate our method, we built threeTSD datasets based on UrbanSound and Audioset. Experimental results show thatour methods offer about 8\% improvement in event-based F-score as compared witha recent baseline.

A Mixed supervised Learning Framework for Target Sound Detection