[New Low-Module Training] Ai Ayumi Hamasaki 48K RVC model

The model is trained using Miaoyin New Low Model, using fewer datasets.

This model is intended only for non-commercial learning and research purposes, encouraging research (AI) development. It must not be used in violation of relevant local regulations.

This model is trained using Ayumi Hamasaki speaking (3 minutes) + singing (4 minutes), with the singing part separated to include harmony and noise.

Here's a brief introduction to our new base model (which will be gradually available to permanent alchemists). If you need to train models, this model should be very helpful to you.

The new base model is retrained based on the base mold of RVC v2, including the original RVC VCTK data. The new base model was roughly trained on the following data.

Chinese: ≈ 11 hours

Japanese: ≈ 14 hours

English: ≈ 5 hours

Cantonese: ≈ 2 hours

Singing: ≈ 3 hours

………….

The total time is about

40 hours of multiplayer conversation data. More details on the low film will be available for download later to update.

Here are some of the tests:

"Plastic Love" - Ai Ayumi Hamasaki

<!--[if lt IE 9]>document.createElement('audio');<![endif]-->

"Piecing Together the Memories for You" - Ayumi Hamasaki

"Red High Heels" - Ayumi Hamasaki

The above is an inference model trained under new low-poly models, and here I also share for free with well-trained inference models for everyone to study and learn.