Saul's Log - November 20, 2001
Recently, Electrode was asked on the forums how I go about
finding loop points in samples. I figure my response might be
interesting or helpful to other musicians, so I'm posting it
here.
I find all of my loop points manually. Though I've
seen some okay software for finding points, I'm not going to
load up a seperate program if I'll end up doing it manually
30% of the time anyway. :P
Generally, if the sample is a drumloop, finding the point isn't difficult, since the impact of each "hit" is easily seen in the wave.
As for melodic samples, I've developed a sort of procedure I follow:
First, record a long sample of the instrument, about 3 or 4 times the length you'll want for the final sample.
What you're eventually looking for is a region that will be appropriate to cross-fade. Generally you don't want a region that has any "anomalies" in it, such as temporary drops/jumps in pitch, or changes in tone quality. You want something that has a steady quality all the way through, otherwise it'll be too easy to hear the loop in the final sample (the sample will sort of "crest" at the loop or in the middle).
If the instrument's attack is important to you (for something like a trumpet), then you need the initial part of the recording (attack + a section for cross-fading) to be anomaly-free. If it's not anomaly-free, you'll have to re-record, get a new sample, or live with it.
If the attack is unimportant your job is easier - just search through the long recording for the steadiest section you can find.
Then trim the sample down to this section, and apply crossfading. There are a few ways to do this.
For a sample with no attack, there's method "A" and method "B". (A): trim to a region double the size of your final sample. Copy the first half of this region and use it to cross-fade the second half. The result is a sample that loops at the ends, and is constantly cross-fading, which makes the cross-fading sound like part of the sample's overall character (as if it's an LFO or something).
(B): trim to a region the size of your final sample plus the size of cross-fade you want tacked on the front. Cut that bit from the front and use it to fade the end of the sample. This again gives you a sample that loops at the end, but this time the cross-fading affects only a portion of the sample.
For a sample with an attack, you can do something similar to method B, except you don't cut out everything before the looping portion. You can include the attack in the cross-fade material (sometimes that sounds fine, and it saves sample space). You'll end up with a sample that loops from the end to somewhere in the middle, with crossfading happening for part or all of the looping portion.
The one time cross-fading can't help you much is with samples that have very complex or dynamic tones. For example, an electric guitar (or other distorted instrument) will often produce a surge of noise in the cross-faded area. However, if the sample is distorted enough, you can sometimes loop without cross-fading.
Generally, if the sample is a drumloop, finding the point isn't difficult, since the impact of each "hit" is easily seen in the wave.
As for melodic samples, I've developed a sort of procedure I follow:
First, record a long sample of the instrument, about 3 or 4 times the length you'll want for the final sample.
What you're eventually looking for is a region that will be appropriate to cross-fade. Generally you don't want a region that has any "anomalies" in it, such as temporary drops/jumps in pitch, or changes in tone quality. You want something that has a steady quality all the way through, otherwise it'll be too easy to hear the loop in the final sample (the sample will sort of "crest" at the loop or in the middle).
If the instrument's attack is important to you (for something like a trumpet), then you need the initial part of the recording (attack + a section for cross-fading) to be anomaly-free. If it's not anomaly-free, you'll have to re-record, get a new sample, or live with it.
If the attack is unimportant your job is easier - just search through the long recording for the steadiest section you can find.
Then trim the sample down to this section, and apply crossfading. There are a few ways to do this.
For a sample with no attack, there's method "A" and method "B". (A): trim to a region double the size of your final sample. Copy the first half of this region and use it to cross-fade the second half. The result is a sample that loops at the ends, and is constantly cross-fading, which makes the cross-fading sound like part of the sample's overall character (as if it's an LFO or something).
(B): trim to a region the size of your final sample plus the size of cross-fade you want tacked on the front. Cut that bit from the front and use it to fade the end of the sample. This again gives you a sample that loops at the end, but this time the cross-fading affects only a portion of the sample.
For a sample with an attack, you can do something similar to method B, except you don't cut out everything before the looping portion. You can include the attack in the cross-fade material (sometimes that sounds fine, and it saves sample space). You'll end up with a sample that loops from the end to somewhere in the middle, with crossfading happening for part or all of the looping portion.
The one time cross-fading can't help you much is with samples that have very complex or dynamic tones. For example, an electric guitar (or other distorted instrument) will often produce a surge of noise in the cross-faded area. However, if the sample is distorted enough, you can sometimes loop without cross-fading.