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

Denoising Diffusion Probabilistic Models (DDPMs) are emerging intext-to-speech (TTS) synthesis because of their strong capability of generatinghigh-fidelity samples. However, their iterative refinement process inhigh-dimensional data space results in slow inference speed, which restrictstheir application in real-time systems. Previous works have explored speedingup by minimizing the number of inference steps but at the cost of samplequality. In this work, to improve the inference speed for DDPM-based TTS modelwhile achieving high sample quality, we propose ResGrad, a lightweightdiffusion model which learns to refine the output spectrogram of an existingTTS model (e.g., FastSpeech 2) by predicting the residual between the modeloutput and the corresponding ground-truth speech. ResGrad has severaladvantages: 1) Compare with other acceleration methods for DDPM which need tosynthesize speech from scratch, ResGrad reduces the complexity of task bychanging the generation target from ground-truth mel-spectrogram to theresidual, resulting into a more lightweight model and thus a smaller real-timefactor. 2) ResGrad is employed in the inference process of the existing TTSmodel in a plug-and-play way, without re-training this model. We verify ResGradon the single-speaker dataset LJSpeech and two more challenging datasets withmultiple speakers (LibriTTS) and high sampling rate (VCTK). Experimentalresults show that in comparison with other speed-up methods of DDPMs: 1)ResGrad achieves better sample quality with the same inference speed measuredby real-time factor; 2) with similar speech quality, ResGrad synthesizes speechfaster than baseline methods by more than 10 times. Audio samples are availableat https://resgrad1.github.io/.

ResGrad: Residual Denoising Diffusion Probabilistic Models for Text to Speech