As of the time of this article on 2024/01/22, the highest ranked on the aicrowd leaderboard is still the non-open source SAMI developed by ByteDance.
If you feel that the current open-source MVSEP-MDX23 and demucs are not very effective, you can try the closed-source SAMI developed by ByteDance's team for source separation.
You can also use the better-performing Duanzi AI, but the price is a bit scary, so this article will not introduce it.
If you encounter poor sound quality, accompaniment overpowering vocals, electronic music, or major audio repairs, there is currently no way to achieve good results.
If you encounter any of these situations, don't try anymore, there won't be good results.
Although I wrote this, I won't listen to advice
Then open the official website and check the price... it's a bit expensive...
If you want to train an AI singer, it will cost hundreds.
No, this is too scary,
not only can't make money, but also have to spend so much
Download a professional version of Jianying.
Click on "Start Creating".
Then drag the audio onto the track, right-click on the audio, and you will see
Then you will find that exporting requires VIP.
The price is not very cheap, but much cheaper compared to buying the API directly.
I Want to Freeload#
Since previewing is possible without VIP...
You will find that previewing is free, while exporting requires payment.
This means you can use any screen recording software that supports internal recording.
But it will take twice the time to extract the accompaniment and vocals, and it requires manual supervision.
Since it is a video editing software, it must have cached the files somewhere.
The address is as follows:
But the file names have become timestamps, which is not very convenient, so you need to rename them yourself.