Messages
This page shows the samples in the paper "Boosting Fast and High-Quality Speech Synthesis with Linear Diffusion".
Experiments were based on LJSpeech and LibriTTS.
The framework of LinDiff:
Single Speaker (LJSpeech Dataset)
| GT | WaveNet | WaveGlow | HIFI-GAN | WaveGrad(6) | FastDiff(4) | LinDiff(1) | LinDiff(3) | LinDiff(100) |
|---|---|---|---|---|---|---|---|---|
Multi Speakers (LibriTTS Dataset)
| GT | WaveNet | WaveGlow | HIFI-GAN | WaveGrad(6) | FastDiff(4) | LinDiff(1) | LinDiff(3) | LinDiff(100) |
|---|---|---|---|---|---|---|---|---|