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 proposes a non-autoregressive extension of our previously proposedsequence-to-sequence (S2S) model-based voice conversion (VC) methods. S2Smodel-based VC methods have attracted particular attention in recent years fortheir flexibility in converting not only the voice identity but also the pitchcontour and local duration of input speech, thanks to the ability of theencoder-decoder architecture with the attention mechanism. However, one of theobstacles to making these methods work in real-time is the autoregressive (AR)structure. To overcome this obstacle, we develop a method to obtain a modelthat is free from an AR structure and behaves similarly to the original S2Smodels, based on a teacher-student learning framework. In our method, called"FastS2S-VC", the student model consists of encoder, decoder, and attentionpredictor. The attention predictor learns to predict attention distributionssolely from source speech along with a target class index with the guidance ofthose predicted by the teacher model from both source and target speech. Thanksto this structure, the model is freed from an AR structure and allows forparallelization. Furthermore, we show that FastS2S-VC is suitable for real-timeimplementation based on a sliding-window approach, and describe how to make itrun in real-time. Through speaker-identity and emotional-expression conversionexperiments, we confirmed that FastS2S-VC was able to speed up the conversionprocess by 70 to 100 times compared to the original AR-type S2S-VC methods,without significantly degrading the audio quality and similarity to targetspeech. We also confirmed that the real-time version of FastS2S-VC can be runwith a latency of 32 ms when run on a GPU.

FastS2S-VC: Streaming Non-Autoregressive Sequence-to-Sequence Voice Conversion