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/

Zendy Plus

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

Zendy Tools

Zendy Open

Text-to-speech systems recently achieved almost indistinguishable qualityfrom human speech. However, the prosody of those systems is generally flatterthan natural speech, producing samples with low expressiveness. Disentanglementof speaker id and prosody is crucial in text-to-speech systems to improve onnaturalness and produce more variable syntheses. This paper proposes a newneural text-to-speech model that approaches the disentanglement problem byconditioning a Tacotron2-like architecture on flow-normalized speakerembeddings, and by substituting the reference encoder with a new learned latentdistribution responsible for modeling the intra-sentence variability due to theprosody. By removing the reference encoder dependency, the speaker-leakageproblem typically happening in this kind of systems disappears, producing moredistinctive syntheses at inference time. The new model achieves significantlyhigher prosody variance than the baseline in a set of quantitative prosodyfeatures, as well as higher speaker distinctiveness, without decreasing thespeaker intelligibility. Finally, we observe that the normalized speakerembeddings enable much richer speaker interpolations, substantially improvingthe distinctiveness of the new interpolated speakers.

Improving multi-speaker TTS prosody variance with a residual encoder and normalizing flows