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

This paper focuses on leveraging deep representation learning (DRL) forspeech enhancement (SE). In general, the performance of the deep neural network(DNN) is heavily dependent on the learning of data representation. However, theDRL's importance is often ignored in many DNN-based SE algorithms. To obtain ahigher quality enhanced speech, we propose a two-stage DRL-based SE methodthrough adversarial training. In the first stage, we disentangle differentlatent variables because disentangled representations can help DNN generate abetter enhanced speech. Specifically, we use the $\beta$-variationalautoencoder (VAE) algorithm to obtain the speech and noise posteriorestimations and related representations from the observed signal. However,since the posteriors and representations are intractable and we can only applya conditional assumption to estimate them, it is difficult to ensure that theseestimations are always pretty accurate, which may potentially degrade the finalaccuracy of the signal estimation. To further improve the quality of enhancedspeech, in the second stage, we introduce adversarial training to reduce theeffect of the inaccurate posterior towards signal reconstruction and improvethe signal estimation accuracy, making our algorithm more robust for thepotentially inaccurate posterior estimations. As a result, better SEperformance can be achieved. The experimental results indicate that theproposed strategy can help similar DNN-based SE algorithms achieve highershort-time objective intelligibility (STOI), perceptual evaluation of speechquality (PESQ), and scale-invariant signal-to-distortion ratio (SI-SDR) scores.Moreover, the proposed algorithm can also outperform recent competitive SEalgorithms.

A Two-Stage Deep Representation Learning-Based Speech Enhancement Method Using Variational Autoencoder and Adversarial Training