The viral AI Stefanie Sun cover songs were made using the So-VITS-SVC AI SVC software. The tutorial is not complicated; the voice training process is relatively time-consuming, and a GPU configuration with 8G+ memory is recommended.
(1). Download the latest So-VITS-SVC one-click installer
1. Mega cloud drive download: 【Click to download】
2. Baidu Netdisk: 【Click to download 】Extraction code: p1bx
GitHub open source
(2). After downloading, first install UVR5, load the model in the archive, put it into models, then perform vocal separation to remove harmonies and reverb; see the Zero Degree video tutorial for specific steps.
(3). Prepare your own clean vocals, record in a quiet environment, preferably no less than 30 minutes; audio file format WAV 【Batch file renaming tool】
(4). Use Audio Slicer to split and slice the audio; each segment should preferably not exceed 7 seconds; finally put the sliced audio folder under dataset_raw in the root directory.
(5). In the so-vits-svc folder, open: 启动webui.bat to officially start training.
(6). After training, open: 启动tensorboard.bat to audition the result; if it is okay, proceed to the final step, synthesis! See the tutorial for operation details.
For better results, it is recommended to train at least 30,000 steps. |