Hey there, it’s been a while since I’ve written any useful articles on my site, it’s time to rectify that. Let’s talk about AI Covers.

Examples of work:

Some glitches are possible, because I don’t have access to the original vocal track, in some tracks there are adlibs that neural network perceives as lead vocals and sometimes there is a mess.

The main question now is, what is needed to make it that way?

Well, first you need a fairly powerful computer, preferably with a video card nvidia 20 or 40 series (with CUDA about 6GB would be OK), you can of course use the processing power of the processor, but it is many times faster. For mac users, both intel and silicon options are available!

To work we need the Replay program

Do I need any plug-ins?

Basically, everything is ready, replay can extract vocals, remove reverb and delays from it, and glue the beat to the vocal, but for more flexibility you can use another great program Ultimate Vocal Remover, or use online services without loading your system.

Do you want to process everything in your DAW? I use Autotune (Waves or Antares), Reverb (Neoverb or Crystalline [best reverb plugin]), any Waves or Nectar de-esser from Izotope, optional – ozone on master in auto processing mode for each track (don’t forget to put vocals on mixer track).

Operating principle

I strongly advise you to move the program from the system disk to the most free before you start.

We take any track that in your opinion can fit to this or that voice, cut the vocal and minus, throw the vocal in Replay, select and download the desired model (on the tab Verified quality models), if we make a male vocal female, add pitch within 10, otherwise pitch down. Default settings are quite adequate, there is a detailed guide on the introduction when opening the software.
Or take vocals from any pack/kit and just run them through the algorithm.

Export, save, go to FL, for example.
2 tracks, throw minus on the first, vocal on the second.
My chain looks like this (depending on the material, one or another plugin fits better). We find out bpm and tonality of your minus and vocal, set everything in autotune, bpm is better to set under the track, because some reverbs sound differently based on the tempo in DAW.

On the master only ozone with auto-tuning, you can skip this step. Straight to the question of how to customize the ozone, just press the circled button while playing drop (Loudes part of track) for better results.

Tips

Cut off extra parts of a cappella with quietness (noise) and use bleeding mod with smoothnes.

If it sounds quiet, crank up the boost with clip.

REMEMBER! Always use quality materials, namely a quality model that was trained on session recordings, otherwise the final work will come out worse than it could be, use vocals without processing!

If you have any difficulties while using replay, ask for help on their server in Discord – https://discord.gg/A5rgNwDRd4

That’s all, use this technique for AI Covers.