This article was first published on the official account Breakthrough an Eagle , and reprinting without permission is prohibited.
I just used AI Stefanie Sun to cover the song "Exchanging the Rest of My Life", and I ran through the whole process, so I will record it.
You can listen to it first:
https://www.bilibili.com/video/BV1az4y1a7K7/
The following tutorial starts:
1.
Find a song to cover, for example, I used "Exchange for the Rest of My Life" originally sung by JJ Lin.
I stepped on a pit here, which caused the final effect to be not very good. As a reminder, if you want AI girls to cover boys' songs, it is best to find the audio of a real female singer's cover, because the pitch of boys is different from that of girls, and the pitch of the accompaniment is also different.
2.
Work with raw audio, separating vocals and accompaniments. A separate software UVR is required here. All the software and material download addresses used in this article can be obtained by replying to "Sun Yanzi" on the first issue.
I have tried two softwares in this step: one is SpleeterGUI, the effect of separating my song is not good, and the labor saving is not clean enough. Later I switched to UVR and it worked fine. I will give you both software, UVR is preferred, if your audio is not good with UVR, you can try SpleeterGUI.
After installation, you need to copy the model file decompressed from "UVR5 model.zip" to the models folder in the UVR installation directory. The default installation location of UVR is:C:\Users\<你的用户名>\AppData\Local\Programs\Ultimate Vocal Remover
3.
UVR is processed for the first time, and the settings are shown in the following figure:
The default output file type is wav, do not modify it. If your computer does not have a graphics card, you can remove the GPU Conversion option.
After the separation is successful, there will be two files in the directory:
In parentheses are Vocals for vocals (dry voice), and Instrument for accompaniment. At this time, you can listen to the effect of the dry sound file. If the reverberation is too heavy, you can go to step 4.
4、
[Optional] If the dry sound reverberation is too heavy, you can use this configuration to process the dry sound again, see the picture below.
5、
Open the so-vits-svc folder in the package, and open the following 启动webui.batfiles. Then a black box of the command line will pop up, wait for a moment. . .
The browser will automatically pop up to this page:
6、
I have put the trained AI Stefanie Sun model in the corresponding directory. The download address of all the software and materials used in this article can be obtained by replying to "Sun Yanzi" on the first issue.
Select the relevant parameters to load the model as shown below:
The image below is a success!
If you need to use other trained models, you usually need to put the .pt, .pth files in the so-vits-svc/pre_trained_model folder, and put config.json in the so-vits-svc/configs folder.
7、
Upload the dry sound (human voice) file, click "Audio Converter" to start the conversion.
The F0 average filter option mainly solves the mute problem. If your output sound has obvious mute, you can check it and try it; the pitch change
can modify the pitch of the output sound (if you don’t understand it, don’t change it)
For example, I asked AI Stefanie Sun to sing the original tone of JJ Lin's "Exchange for the Rest of My Life", which feels a bit low. You can raise the tone. I set the tone change to 5 (equivalent to going from the original tone D to G). But you need to pay attention to the rising key of the accompaniment, otherwise it will sound strange after synthesis.
8、
Wait for a while, after the conversion is completed, the following picture will be displayed, you can play the trial listening, click the three points to download:
9、
Synthesize dry voices and accompaniments. At this point we already had the dry cover and the original accompaniment. You can use clipping to synthesize dry sound and accompaniment, and then output video or pure audio, so I won’t talk about it here.
I raised the key during the cover conversion, so the accompaniment also needs to be modified, otherwise the vocals will not be in key. I use Audacity (https://www.audacityteam.org/) to raise the accompaniment, drag the accompaniment into it, press Ctrl+A to select all, select "Effect > Change Pitch" in the menu bar, and adjust it as needed. Do not save directly after the adjustment, directly save the audacity project file, and select "File > Export" in the menu bar.
Summarize
This article is the process of using an existing model to perform a cover song. If there are more interested people, I will write a tutorial on training AI singers from 0.
Thanks to @归宗@重车者不鸟@风布団 several bigwigs for their information and tutorials.
Love landing on the planet does not talk about imaginary, but only talks about how AI lands into productivity and money!
If you want to know:
- How to use AI to increase work efficiency by 10 times and improve workplace competitiveness
- How to Use AI to Engage in a Sideline Business and Earn Millions a Year
- How to use AI to improve life happiness index
- How peers are using AI to reduce costs and increase efficiency
Welcome to chat with me.