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

Variational auto-encoder (VAE) is an effective neural network architecture todisentangle a speech utterance into speaker identity and linguistic contentlatent embeddings, then generate an utterance for a target speaker from that ofa source speaker. This is possible by concatenating the identity embedding ofthe target speaker and the content embedding of the source speaker uttering adesired sentence. In this work, we propose to improve VAE models withself-attention and structural regularization (RGSM). Specifically, we found asuitable location of VAE's decoder to add a self-attention layer forincorporating non-local information in generating a converted utterance andhiding the source speaker's identity. We applied relaxed group-wise splittingmethod (RGSM) to regularize network weights and remarkably enhancegeneralization performance.  In experiments of zero-shot many-to-many voice conversion task on VCTK dataset, with the self-attention layer and relaxed group-wise splitting method, ourmodel achieves a gain of speaker classification accuracy on unseen speakers by28.3\% while slightly improved conversion voice quality in terms of MOSNetscores. Our encouraging findings point to future research on integrating morevariety of attention structures in VAE framework while controlling model sizeand overfitting for advancing zero-shot many-to-many voice conversions.

Enhancing Zero-Shot Many to Many Voice Conversion with Self-Attention VAE