If the song has heavy stereo reverb on the voice (common in shoegaze or 80s ballads), you are doomed. The reverb is spread to the sides, so when you cancel the center, you lose the voice but keep the echo. You end up with a ghost singing from a well.
For decades, this was impossible. A finished stereo mix was considered a "brick wall"—you couldn't pull the bricks out without breaking the wall. vocal isolation audacity
You highlight a section of music. The AI analyzes the waveform and asks, "Does this frequency pattern match a human larynx or a cymbal crash?" It then tries to erase the non-voice parts. If the song has heavy stereo reverb on
It’s too good. If you isolate the vocals from a Queen song, you’ll hear Freddie Mercury in your room. But listen closely: the AI sometimes eats the guitar solo that was harmonizing with the voice. Or it leaves behind "digital butterflies"—shimmering, ghostly artifacts that sound like a choir of robots. The Secret Sauce: Embrace the Wreckage Here is where most people give up. They isolate the vocal, hear the artifacts, and delete the file. That is a mistake. For decades, this was impossible
This produces shockingly clean a cappellas. You can often hear breaths, lip smacks, and room reverb that were buried in the original mix.