Razer112/DMR_Pretrain

DMR: Deep and Soft Voice Improvement Pretrain for RVC

Model description

DMR is a pretrain designed to improve soft and deep voices.

Intended uses

• Improve voice conversion quality, especially for soft and deep voices
• Enhance breathing sounds (and eventually whispering) in voice conversion
• Make better E-girl and Mommy voices :3

Training data

The model was trained on a custom dataset with the following details:

• Total duration: 11.3 hours
• Language: English
• Number of speakers: 22
  • 16 female speakers
  • 6 male speakers

Training Process

DMR was trained with Applio using a RTX 4060TI 16gb.

• BatchSize: 8
• Pitch Extraction Method: Mangio-Crepe
• Hop Length: 32
• Sample Rate: 32K

Usage

To use the DMR pretrain:

1. Download both the D and G files of the DMR model.
2. For standard RVC setup:
   • Place the downloaded files in the pretrained_v2 folder.
3. For Applio users:
   • Place the downloaded files in the custom pretrains folder.

Additional Information

I do plan to make a V2 of DMR with around 40 hours of speech using a newer GAN but I do not have a release date.