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

The cross-speaker emotion transfer task in text-to-speech (TTS) synthesisparticularly aims to synthesize speech for a target speaker with the emotiontransferred from reference speech recorded by another (source) speaker. Duringthe emotion transfer process, the identity information of the source speakercould also affect the synthesized results, resulting in the issue of speakerleakage. This paper proposes a new method with the aim to synthesizecontrollable emotional expressive speech and meanwhile maintain the targetspeaker's identity in the cross-speaker emotion TTS task. The proposed methodis a Tacotron2-based framework with emotion embedding as the conditioningvariable to provide emotion information. Two emotion disentangling modules arecontained in our method to 1) get speaker-irrelevant and emotion-discriminativeembedding, and 2) explicitly constrain the emotion and speaker identity ofsynthetic speech to be that as expected. Moreover, we present an intuitivemethod to control the emotion strength in the synthetic speech for the targetspeaker. Specifically, the learned emotion embedding is adjusted with aflexible scalar value, which allows controlling the emotion strength conveyedby the embedding. Extensive experiments have been conducted on a Mandarindisjoint corpus, and the results demonstrate that the proposed method is ableto synthesize reasonable emotional speech for the target speaker. Compared tothe state-of-the-art reference embedding learned methods, our method gets thebest performance on the cross-speaker emotion transfer task, indicating thatour method achieves the new state-of-the-art performance on learning thespeaker-irrelevant emotion embedding. Furthermore, the strength ranking testand pitch trajectories plots demonstrate that the proposed method caneffectively control the emotion strength, leading to prosody-diverse syntheticspeech.

Cross-speaker emotion disentangling and transfer for end-to-end speech synthesis