Such as when people make “isolated bass/vocal,” tracks of popular songs or make a meme mashup of one song with he vocal line of another over it?
In: Technology
In addition to isolating frequencies and stereo, there are also a few other ways.
1: Isolate and chop different parts of a beat to isolate each instrument, you can pitch up or down if you only get one note.
2: If you need an acapella, and have the original track and the instrumental, you can sometimes isolate the accapella by combining the original with a version of the instrumental with the waveforms inversed to cancel each other out
To add to what the other commenter said, audio files are usually in stereo – there are actually 2 tracks, one for the left side and one for the right – and when a song is created, different sounds are usually placed in different locations in the stereo field. Lead vocals, kick drums, bass, and sometimes snare drums are usually placed in mono – the left and right channels are identical. Most other sounds are usually not in mono – the left and right channels are different. For example, a sound may be panned, where the volume on one side is louder than the other.
If your goal is to extract vocals, you can simply discard all the data that isn’t the same on both the left and right channels. You will be left with less overlapping frequency information to try and cut the vocals out of.
There are also AI tools that can assist with separating tracks, such as Spleeter.
If the source is a stereo 2 track, then the answer is that much like a baked cake, you can’t extract just the eggs out, or just the flower out. You can however, try and filter out certain frequencies to try and reduce the other tracked items in the mix, but it doesn’t sound very good. Remixes often times use the source instrumentals or vocal tracks as well, it’s not uncommon for that to be done.
While EQing can help, you’re correct that there is far too much overlap for this method to be really successful on its own unless you’re looking for something that sits in its own range of a song that’s mixed well, like bass. For vocals, the most common practice relies on how most vocal music is mixed; vocals down the center, instruments panned to the sides. By converting a stereo track to left and right mono signals, phase inverting one side, then playing them back together, you’ll cancel out everything down the middle; typically just the lead vocal. By inverting an instrumental track over the original full mix, you can isolate a vocal. There’s many tutorials out there on YouTube, and it’s definitely a fun way to play with audio!
There are multitrack projects of a lot of songs, some available to the general public if you know where to look for them, and some only accesible to people in the recording industry with *contacts*. If you have one of those projects, you can just mute the track you don’t want (vocals to karaoke, or some instrument if you want to make a cover, etc)
The song contains a range of sound frequencies. Think of these frequencies as “audio colours.” The bass is red, and all the way on the other end do the rainbow is high-pitched purple.
To isolate just one voice in the music, you can filter the spectrum so you only have the range you want — just the “orange”, for example, or just the “green”, like screwing filters over a lens, where some colours are blocked and others are let through.
With software it’s relatively easy to choose just the exact frequencies that comprise a particular sound in the mix, and filter out the rest. A couple of passes like this with different settings, and you have imperfect but workable “separated” tracks.
————————
Edit to add: I do this in the course of my work, when creating generative visuals from mixed-down audio files.
There are many different approaches. To add to what other folks have said, these days you can use neural networks that have been trained to do the job. [Spleeter](https://github.com/deezer/spleeter) is one such software. Basically, a computer “learns” what different instruments sound like, and can then identify the different parts in the spectrum of the full song, and then cuts out those bits separately. It works well with clear, well defined, loud instruments in simple tracks; not so much with things buried in a complicated mix (where even a human would have a hard time picking them out).
Others have also mentioned that if you have the instrumental track, you can subtract it from the full track to get just the vocals. This works rather well for some tracks, but the two tracks have to be perfectly aligned for that to work directly. There is software that can help with this by automatically stretching and shrinking tracks in time to match them (I wrote a version of this myself). Sometimes it doesn’t work well anyway, because the tracks might have been processed differently or different instruments might be out of phase (this often happens with synthesizers and other virtual instruments, where both versions aren’t the same render to audio). In this case you can use even smarter software to try to subtract at the level of tones in the spectrum, instead of at the level of the raw audio.
Ultimately it’s all a bunch of compromises, and if you’re doing this by hand you’re probably going to be postprocessing stuff manually to cut out noise, EQ the result, etc. It’s not a one button push process.
Not commenting on isolating individual instruments, but a common trick to remove the voice for a karaoke track is to subtract left from right channel. In most stereo recordings, the voice is present with equal intensity in both channels.
This is how cheap karaoke machines work.
Latest Answers