What approaches can be taken to extract or recreate audio for one or more individual parts from a recording with several layered parts?
A number of approaches can be taken to reaching the desired goal, of isolating the audio for individual parts, within multitracked recordings. Different approaches may be appropriate for different styles of music.
If an audio recording only has one instrumental/vocal layer, at any particular point (or only one per stereo channel) it is easy to extract the audio for this part, as it will be the same as the original recording! This is the most effective way to take samples from recordings; it seems obvious, but use a bit of the track that only has the instruments/vocals you want, with no other instruments. (A good example is the sample of Suzanne Vega's Tom's Diner used for this remix.) But, of course, the audio you want may already be part of a multi-layered audio texture. And, once multiple instrumental/vocal tracks have been mixed down to two stereo channels, isolating one individual part becomes far more problematic. In this case, three main approaches will be useful:
- It is possible to use filters to remove certain frequencies from the audio. This answer to this question addresses this approach. This will be most effective if the audio you want to extract is largely in a different band of frequencies to the other parts. (It should be reasonably easy to extract the audio for the tuba part from a tuba and flute duet recording!) However, the range of frequencies in the audio of any one instrumental/vocal part are often wider than one would expect (once transients, overtones and other resonance are taken into account, for instance). Usually it is only possible to boost some instrumental/vocal parts and cut others; it is difficult to completely isolate parts from a mixed recording. In fact, there are various types of software for doing this. This question, and it's answers, discuss cutting the frequencies of the guitar part in a recording (okay, it's kind of the opposite of isolating a track, but the same principal!)
- It's possible the audio you want to extract from a recording is already a sample from elsewhere. This is most likely if the audio is part of a recording in a genre that tends to use sampling (for example, Hip-Hop, but now used in a wide variety of popular music.) There is no exact way to find out if the part you want within an audio recording is a sample, but there will be plenty of clues: if the audio sounds significantly different in style to the other audio layers (eg. "orchestral-sounding" music within a rap song); if the section of audio returns multiple times within the song/piece sounding exactly the same. But these are only clues, and may not be indicative of a sample, after all: the audio may have been recorded for the song/piece in question, even if it seems to be different in style; it is common to repeat small sections of audio, recorded in the studio, in a song/piece - this is effectively the same as creating your own sample. There are other ways to find out if a recording uses sampled audio from elsewhere, though. Websites such as www.whosampled.com have extensive lists of sources of samples used in recordings; the Wikipedia entries for songs often have information about samples used, too. If the audio you want to isolate from a recording is a sample from another recording, you can simply seek out the original recording to find the source of the sample, and so the individual audio part.
- If neither of the approaches above works, you can recreate the audio yourself (yes, this is a much harder approach!) Listen to the audio within the recording, that you wish to isolate, and work out what notes it uses. This process is called transcription. There are various types of software available to aid this process, for instance Transcribe! Once you have worked out the notes in a section of an audio layer, you can try to recreate this audio as an individual audio recording, either with live instrument(s) or sampled instruments. Of course, it may not be possible to fully recreate the tone, inflexion and other characteristics (eg. applied effects) of the original audio. As I say, this is by far the most difficult way to get the audio part you want from a recording, but it is still likely to be easier than getting in touch with the artist, band or record company and being allowed access to the original multitrack recordings!