48K Ai Yalin A mature RVC model with a wide vocal range

vocal range performs well in AI covers. Since the platform can only upload limited preview effects, I played two songs with different vocal ranges during the singing demo, which I could listen to in full.

The model is processed similarly to previously released models, with gradual denoising of the original dataset, removing almost all noisy areas in the dataset.

In terms of inference, I haven't encountered any distortion or breath cracking sounds so far. There may be rare cases where this issue occurs. I've tried it so far and found that some original input can output noiselessly.

As for laughter, most of the dataset related to this area was fed into the training, and the output was decent. Just like with the previous model of Limo, everyone's laughter varies from person to person, so the effect may vary depending on how you use it.

Most so-called models that can "laugh" are actually fixed, because to think of an adversarial generative AI like RVC, you need to imagine it as a "teacher" and a "student." The "student" only learns certain specific languages. To support diverse smiles, it needs to be taught many different smiles, which obviously requires a large dataset on tone expression.

The model's range is roughly between -5 and -6 in the original mature tone tone, which is the result I tested later.

Breathing sounds and such are fine, with no "zi" sound, and the song's reasoning is also well presented.

Effects vary from person to person; it doesn't mean your output will be exactly what I demonstrate.

I uploaded a copy from Google Cloud, and you can also download it from Google.