utautautau / pjs-diff-svc

PJS Diff-SVC model public release

pjs-diff-svc

This repository was made for a public download of the Diff-SVC model I have trained using the PJS Corpus. This model is licensed with the CC BY-SA 4.0 license.

This database is a masculine sounding voicebank. It has 27 minutes and 51 seconds of data. The specific range of this database is from G2 to B4. This model was pretrained over Gakuto Deshimaru Diff-SVC. The pretrain model was trained over 2 hours of data, up to 70k steps.

For more information about the PJS corpus, please visit this webpage. I would like to thank Junya Koguchi and Shinnosuke Takamichi for making this database publicly available.

Example renders

Reference audio is a Synthesizer V Pro render with Yuma using Instant Mode tuning.

pjs_oborozukiyo.mp4

Reference audio is a Synthesizer V Pro render with SOLARIA using Instant Mode tuning.

pjs_daisy_bell.mp4