I was recently approached by a couple of CD-ROM houses to come up with short loops of my own ambient music. My remit was to produce loops of no more than 16 seconds that would cycle seamlessly, with no obvious start or stop points. The idea was that when the loops were triggered by users entering a particular environment in the CD-ROM, the users would be encouraged to stay there, rather than leave as fast as possible to escape an obviously repetitive and irritating musical loop. I had two options for making the loops: one was to use my Ensoniq EPS16+ sampler, and the other was to use my Soundscape hard disk recorder.
For some people, looping on samplers comes as second nature, but for many, the results can be very hit and miss. Looping simple drum grooves usually presents no problem, but what about sounds that evolve over long periods, where the timbre and volume at the end of the sample can be radically different from the start? Conventional looping (whereby a section of the sample is looped to conserve memory after the initial attack of the sound has passed) is not appropriate here. The aim is to provide a loop that uses the entire sample and cycles to infinity without sounding like a broken record -- so some sort of crossfading between the beginning and end of the sample is called for. Although my trusty Ensoniq has various crossfade options, the parameters are entered numerically, not graphically, making for a potentially long-winded trial and error process. With a tight deadline, I needed a process that was quick, efficient, and guaranteed to work every time. Enter Soundscape.
Although Soundscape is my system of choice, you should be able to adapt the instructions below to all the current computer-based hard disk systems (and with a little lateral thinking, to the stand-alone units too), since they all have the tools to do the job.
Once you have loaded the source material onto your hard disk, the first thing to do is locate the region you would like to loop and then cut around it, giving yourself a few extra seconds either side. Make sure that the time axis of your display is measuring bars and beats, as opposed to SMPTE time. For textures, the actual tempo and time signature is not critical -- the bar lines merely serve as an accurate scale to help you cut segments of equal length. In my case, 120bpm in 4/4 was the optimum setting, since eight bars then gives exactly 16 seconds, the length of my required loop.
If the loop start point is critical, locate it and make a cut here. Now set the snap value for your move tools to 'bar' and move the cut segment to the start of a bar line, so that your sample starts exactly on the bar. Figure 1 (left) shows the source material cut in this way, with the loop start moved to the beginning of bar 9.
Now turn the snap value to 'off', move the section preceding the cut right up to the cut, and glue the two parts back together. Switch the snap value back to 'bar', and indicate the segment you wish to loop (in my case, eight bars) with left and right markers. If there is no specific place you'd like the loop to start, there is no need to perform the cut and move operation -- simply place markers on the bar lines that delineate the loop.
Here's where the fun starts: you need to create a linear fade-in from silence to full volume near the beginning of the section you wish to loop, and a linear fade-out from full volume to silence near the end of the section. How rapid you make these fades depends on the nature of the material you are looping, as we shall see later (see 'Fine Tuning' below). However, there are two important points to note: firstly, the fades must be of the same duration at either end; and secondly, the start and end of the section you wish to loop must come exactly halfway through the fades you make, however long you decide the fades should be. You first cut the audio up into three chunks, so that you can impose the fades in the chunks at either end without affecting the middle section. (In my example, I decided the fades should be two bars long at either end, and therefore cut the audio into the following three parts: a 2-bar section at the beginning, cut so that the start of the eight bars I wanted to loop was halfway through this chunk; a 6-bar section in the middle; and a 2-bar section at the end, with the end of the eight bars I wanted to loop falling in the middle of this third chunk). You then place a linear fade-in on the first chunk, leave the central chunk as it is, and place a linear fade-out over the third chunk.
Copy the set of all three chunks onto another track, and move the copied set of chunks so that its fade-in chunk exactly overlaps the fade-out chunk of the original set (remember, the fade-in has to be the same length as the fade-out, so this shoudn't be too difficult). Repeat the process a third time, so that the result looks something like the top section of Figure 2 (above). You now have three sets of three chunks each, and each set crossfades into the next. At this point, you can audition the crossfades to see if there's a smooth, inaudible transition from the end of one set of the three chunks to the start of the next.
If you're happy with the result, place the left and right markers so that the left marker falls halfway through the fade-in on the second set of the three chunks, and the right marker falls halfway through the fade-out on the same set. In other words, the markers should delineate the section you wish to loop in the middle set of the three chunks. This sounds confusing, but don't panic -- just take a look at Figure 2. The middle set of three chunks has been copied to another track, and all three chunks are marked 'Copy 1'. As described above, the dotted marker lines are positioned exactly halfway through Copy 1's fade-in and fade-out, giving in this case eight bars between the markers; the length of loop I wanted.
Providing all the pan, volume and EQ settings are the same for both tracks, the way the material has been copied ensures that the audio to the right of the left marker sounds exactly the same as everything to the right of the right locator. So, if you now digitally merge the material from both tracks between the two markers, you will have a section that will play seamlessly to a copy of itself -- since the beginning is an exact copy of what followed the end. In other words, it will loop perfectly.
All that remains now is to export the section between the markers to your sampler. There are a couple of ways of achieving this. If you have a sample editor such as Recycle or SampleCell II, save your loop on your computer's hard disk as a WAV file (on PCs) or an AIFF file (on a Mac), import it into your sample editor, and do a standard sample dump via SCSI to your sampler. The new Akai samplers have their own sample editing software for the Mac (and soon for PC), MESA. This reads AIFF files, which can then be dumped to the sampler. Alternatively, simply record the merged section between the markers onto your sampler, then truncate it with the aid of your waveform display, so that there is no dead space either side of the sample, and set the loop start and end points to the start and end of the sample. If your sampler has no display, you will have to scrub through the sample numbers to listen out for the start and stop points before truncating.
As hinted earlier, there is no reason why the fade-in/fade-out chunks need to be two bars long, as in my example; they can be any length you like, provided they are equal. Depending on the material you are trying to loop, if the transition between loops sounds too abrupt, make the fades longer, ensuring above all that you keep the left and right markers in the middle of their respective fades. The longer the fade chunks are, the more material either side of the left and right markers will be incorporated into the final sample, so the resulting loop will sound less like the original source material. If the volume still seems uneven across the loops, try experimenting with different fade-in/out curves, using a convex curve for the fade-in and its inverse shape on the fade-out, for example.
You may wish to loop a section of music which is not a whole number of seconds long, so using 120bpm as a grid will not give an accurate stop point. By increasing the bpm of your time axis, you can home in on the source material with more accuracy, since you have more bar lines over a given period. Better still, if you have a 'compute tempo' facility, place the markers at the start and end of the section you want to loop, and tell the program that this section is (for example) eight bars long. Next, cut and move the section to the start of one of the new bar lines, re-glue the preceding section as before, and you are now ready to make your bar-accurate cuts.
Once you've got into the swing of making loops in the way I've described, you may find it difficult to go back to looping on the sampler. Besides providing seamless loops for CD-ROM environments, you could use the results to make complex drones and animated ambient washes for your compositions. If you are sampling wholesale from a track, looping in this way will stop your loops from sounding stilted, without the abrupt change as the sample jumps from the end to the beginning of the loop. By using a suitable Java program, you could even set up your web page to automatically trigger and loop your sample, so that every time someone visits your Internet site, they get a memory-friendly taster of your music.
People spend years perfecting their looping techniques on a sampler, but with this method, your loops need never go near a sampler, and you can obtain perfect results in a couple of minutes.
Using crossfades to loop samples with rhythmic elements is much more hazardous than trying to loop non-rhythmic drones; if you try to loop sections without paying attention to the tempo of the source material, eight bars (say) on the grid will almost certainly not be eight bars of your music. You might get a result that seems to stay in time with itself, but more than likely, it will create a loop with an incomplete bar at the end, or there will be interference at the crossfade, with completely mismatched beats. In this situation, performing a 'compute tempo' operation as described elsewhere in this article (see 'Fine Tuning') is the only solution. Even then, if the rhythm hasn't been played with metronomic accuracy, you may get flams in the crossfades. If this happens, reduce the crossfade size to as little as a crotchet's worth (quarter note). As long as you use a small amount of overlap, you'll still retain some of the sound's natural decay and reverb at the end of the sample, avoiding the abrupt changes that can occur when you make loops without employing crossfades.