In a similar vein to my previous rant about why you can’t convert between MIDI and audio data, I thought I’d address another question I see a lot. This question comes in two forms, but it’s basically the same thing:

• Can I make an A Capella version an MP3 so I can remix the vocals with a different beat?
• Can I remove the vocals from an MP3 so I can make a karaoke version of a song?

The answer to both questions is “no”. More precisely, it’s “technically yes, but not really, and if you are asking this question, then you don’t have access to the data which would give you what you wanted or else you wouldn’t be asking”. Hopefully this guide will explain why.

The reason that both of the above questions are fundamentally the same thing because they involve two opposite processes that are used when working with audio data: mixing and filtering.

Mixing is the process of taking two or more audio streams and combining them to make a single stream. So when a band goes to the studio to make an album, all of the instruments are recorded separately and then mixed down to make a single track. This is done so that each instrument can be processed with effects to improve the sound, and then its volume is adjusted accordingly so that it sounds nice with the other instruments (note: this is a vast simplification of the entire process).

Filtering is the process of removing or boosting particular frequencies of an audio signal. There are lots of different types of filters each designed to do something different. Since we are talking about vocals today, it’s worth noting that the average human voice has a frequency range of 80 Hz - 1100 Hz. (The range of human hearing is approximately 50 Hz - 20,000 Hz). When most people think about removing vocals (or the other instruments) from a song, they think that they can filter them out. Unfortunately, it’s not that easy.

Why not? So the band went to the studio, recorded a song, and mixed it down. Why can you “unmix” it? Well, since the tracks are already mixed together, you would need to remove just the frequencies of the song which contain the singer’s voice (or the other instruments, if you want to make an A Capella version). The problem is, a piano has a frequency range of 27.5 Hz - 4186 Hz. The guitar has a frequency range of 80 Hz - 5000 Hz. And so on.

So basically, you can remove the vocals from a song by filtering them out, but not without removing parts from a bunch of other instruments as well. And precisely because you’d be removing only parts of the other instruments, the resulting song would sound very weird and not at all like the original. So not very much fun to remix or to sing along to. Likewise, if you want to remove all of the other instruments to make the A Capella version, you’ll find that there will always be some traces of the other instruments along with the voice, so it wouldn’t sound very “clean”.

This is simply because a lot of the musical instruments we use occupy the same frequencies as the human voice.

But maybe it’s possible to construct a filter which just takes out the exact frequency of the particular note being sung at a given moment and not the entire vocal range? Unfortunately, no, this is not possible. Doing this by hand would take ages. And machines simply do not hear audio in the same way as humans do. It is not possible to tell a computer to recognize and separate the human voice apart from all the other instruments, just like most people probably couldn’t correctly put the labels on a full box of crayons. And especially if they had to identify which ones were used in a particular drawing.

But, even if you could get a computer to recognize the particular frequency fingerprint of a singer, it still wouldn’t work because once again, parts of the sound from the other instruments will bleed into the singer’s frequency range. And the resulting output, whether you are considering the karaoke version or the A Capella version, would sound weird.

What about all that shareware software which claims to extract vocals from MP3’s or make A Capella versions from your favorite songs? Simply put, it will not work. Save your money and your time, because this software applies a simple filter to the audio data, and you will be disappointed when you hear the result. Trust me on this one. The people that make these things are dishonest, and sometimes simply repackage open-source audio editing software and re-label it as their own.

So I said that the long answer to this question is, “technically yes, but not really, and if you are asking this question, then you don’t have access to the data which would give you what you wanted or else you wouldn’t be asking.” Why is that? Well, the “not really” part is what I have just explained above.

But let’s think back to the mixdown. Say you recorded a band, and you still had the original recordings of each instrument which was used to make the final version of a song. Then you could easily take just the vocal tracks and save them to a separate file to make the A Capella version. Or you could mute the vocals and mix the track with all of the other instruments to make the karaoke version. But either way, if you already had access to the original recording session, then you’d realize that you could make an A Capella or karaoke version if you wanted to, and in that case, you wouldn’t be asking. :)

Questions or comments? Send an email to . Copyright (c) 2012 Teragon Audio. All Rights Reserved.