wenet-e2e / wetts

Production First and Production Ready End-to-End Text-to-Speech Toolkit
Apache License 2.0
356 stars 56 forks source link

[vits] Support WavLM Discriminator #215

Closed Shengqiang-Li closed 3 months ago

Shengqiang-Li commented 3 months ago

Support the wavlm discriminator, which leverages large speech language model representations to enhance the naturalness of the synthesized speech. (see this paper https://arxiv.org/pdf/2306.07691.pdf) 20240330-141112 img_v3_029e_d2073880-71da-47fd-9145-ac7e576a80dg