48k mature female voice, free to stream. 48k mature female voice, free to stream

This batch of models is a Shaoyu Yin reconstructed using SPIN to train a 48k base model.

The currently released model is a pre-trained model that hasn't fully run yet. In testing, we used the same data segment to compare the ContentVec model with SPIN. It does change mutes and treble in certain areas, but since it hasn't run completely, we're not sure about the final effect.

You don't need to change any original RVC to run the inference model.

（This model, which we’re releasing now, features a reduced level of speech control and is trained using a 48k base model reconstructed using SPIN.

SPIN is based on a self-supervised learning framework and employs innovative speaker perturbation and noise invariance training strategies to more effectively decouple speech content from speaker features. Its core advantages are:
1. Training efficiency: Compared to ContentVec, SPIN converges faster and requires fewer computational resources.

2. Representation quality: It performs better in terms of speaker similarity and speech naturalness.

3. Noise robustness: It has better resistance to background noise.

The model currently being released is a pre-trained version that hasn’t been fully run. In testing, we compared the ContentVec model with SPIN using the same dataset. It does indeed alter muffled and high-pitched sounds in some places. Since it hasn’t been fully run, we’re unsure of its final performance.

For the inference model, you don’t need to modify any of the original RVC.）