pc_tgm_hifigan is a fine-tuned pc-nsf-hifigan vocoder. It was fine-tuned off of OpenVPI's pc-nsf-hifigan checkpoint.

This vocoder was fine-tuned off of 132 hours of Singing, Speech, Rap and Stringed data. Please see the "xlsx" file to understand the breakdown of data in this vocoder.

It has been tested with voices included and excluded from the corpus used to train it, and has shown to improve the rendering quality of many voices.

The license is included in this folder, but simply put, you may:

• Use this vocoder on any DiffSinger voice library, so long as it does not break that voices individual TOS.
• Distribute this vocoder with your DiffSinger library for non-commercial use.

You may not:

• Claim this vocoder as your own
• Distribute under a different name
• Generate revenue using this vocoder in any way, as it breaks the CC-BY-NC 4.0 license agreement.

Data credits and amount breakdowns can be found here.