You may want to take this with a grain of salt, because it could be completely wrong. I am familiar with music industry recording, and I am aware of capabilities of computers; I am going to wager an educated guess.
When music is mixed in recording studios, all the music and vocals are kept on separate tracks. that way, if they want to change something, they can change it without having to re-record everybody else's parts (studio time is incredibly expensive). When tracks are made for Karaoke, they are mixes that do not include the vocal track.
On the consumer version of a song, everything is collapsed into two tracks. Computers are good at picking out various frequencies, but it becomes incredibly difficult when everything is mixed together. A human sitting down can manually filter out some of the vocals, but in places where the vocal frequency happens to match the frequency of the music, it is going to filter the music also.
In a nutshell, I would be very skeptical of any software that can claim to accomplish what this is claiming. I would be equally skeptical of any software that could claim to decipher captcha. It is sort of the same concept.
If they offer a trial, give it a whirl and see how it works, but I wouldn't invest any money in it until I had a chance to test it on a variety of music.
And again, I could be completely wrong.
It would only be the umpteenth time today.