A self-training free Ai Xiaorong female RVC model

Wishing everyone a Happy New Year in advance! ～

⚠️ :The model is for non-commercial communication and learning only. The model's sound comes from the internet and does not have the copyright for this sound.

I trained an RVC model myself, which I released during the New Year for everyone to learn and research. The updated model's voice is girly, with a wider range and good support for high notes, meaning it's great for female covers.

Training was only done 200 times, and I used AI noise reduction and randomly filtered some datasets for listening. There might be some noise, but so far I haven't encountered any during inference.

dataset used about 12 minutes of audio and captured most of the high and low frequencies.

The dataset is not copyrighted and can only be used for non-commercial learning and research.

Below are some of the training logs

INFO:mygf-xiaorong:loss_disc=3.787, loss_gen=3.440, loss_fm=11.325,loss_mel=22.061, loss_kl=1.287
INFO:mygf-xiaorong:====> Epoch: 76 [2025-01-23 21:57:44] | (0:00:06.973495)
INFO:mygf-xiaorong:====> Epoch: 77 [2025-01-23 21:57:50] | (0:00:06.382966)
INFO:mygf-xiaorong:====> Epoch: 78 [2025-01-23 21:57:57] | (0:00:06.292653)
INFO:mygf-xiaorong:====> Epoch: 79 [2025-01-23 21:58:03] | (0:00:06.280783)
INFO:mygf-xiaorong:Saving model and optimizer state at epoch 80 to ./logs/mygf-xiaorong/G_2333333.pth
INFO:mygf-xiaorong:Saving model and optimizer state at epoch 80 to ./logs/mygf-xiaorong/D_2333333.pth
INFO:mygf-xiaorong:saving ckpt mygf-xiaorong_e80:Success.