Techniques for making samples loop better

[sheaf]

I'm looking for production techniques that might have been used by engineers in the 80s and 90s to make the samples included with rompler synths loop properly.

If you take the SY77 for example, for some samples the loop is just a single cycle waveform with wavelength that corresponds to the fundamental frequency. If you try this with any random instrument sample from the internet you'll hear the point where it starts looping can sound pretty discontinuous and jarring. Like a sudden timbre shift. They somehow managed to make this sound pretty smooth in most cases.

For other samples the loop is an actual segment of the sample, and often this too sounds really good. If you try to replicate this with a random instrument sample you find online most of the time this will introduce a significant "beating" effect as it loops. The speed of the beating then changes with the note played and this can sound really fisher price.

When creating my own loopable samples I tried to use the following principles:
- The sound is resampled with a sample rate set such that the wavelength corresponding to the fundamental frequency is an integer.
- The loop length is always a multiple of this wavelength.
- Cross fading is optionally applied to make the transition smoother (just like in Kontakt).

With just these techniques I'm not getting results that are good enough, especially with brass and woodwind samples. I tried to smooth out the harmonics with a dynamic EQ to make the overall sound more even, but this barely moves the needle.

The only technique I can think of is to implement a system described by Xavier Serra in his PhD thesis. Basically you resynthesize the sound by splitting it up into a sinusoidal and a residual part. The sinusoids for the harmonics each have an amplitude track that you can easily stabilize by smoothing it out. The residual part is resynthesized using dynamic filtering of noise. These filter coefficients can of course be time smoothed too. Hopefully this will reduce the beating effect?

I don't know how well this is going to work in practice though, and it's a lot of work to properly implement, so I'm hoping there are other techniques I can use with existing software.

Any ideas? Or did they do it simply by having excellent recordings?

[lfm]

Depending on instrument the tone will not be endless, you mention brass and woodwinds, they only last a breath out if maybe 3-4s and can be recorded in full. No loop points needed, depending on the number of articulations made with short stabs or not.

More percussive instruments like piano will probably need to be recorded in full as they fade out.

If doing string section you have when a player change direction of the bow, but not everybody will do that syncroniously, but probably some are syncronous and can spotted and made a good loop point.
- this for natural playing of a partiture live of a piece
- but when recording for sample libraries they will probably be syncronous and one stroke is recorded, or maybe two so a loop point will occur
- good libraries like Spitfire Audio Hans Zimmer strings have loads, and loads of articulations

So careful planning of articulations needed would be a good start before doing recording.

Just a few thoughts....

[RenderFarm]

Try to run the sounds through PaulXstretch (free) which will even out the changes in pitch and volume due to the stretching. You should then be able to loop the samples a lot easier while remaining the characteristics of the sound. You might want to split the transients and the sustain part of the sample, because the attack of the original sample will be smeared in time as well. I have tried this approach myself and it does work for looping, but does introduce phase issues in stereo samples.

[sheaf]

Depending on instrument the tone will not be endless, you mention brass and woodwinds, they only last a breath out if maybe 3-4s and can be recorded in full. No loop points needed, depending on the number of articulations made with short stabs or not.

More percussive instruments like piano will probably need to be recorded in full as they fade out.

If doing string section you have when a player change direction of the bow, but not everybody will do that syncroniously, but probably some are syncronous and can spotted and made a good loop point.
- this for natural playing of a partiture live of a piece
- but when recording for sample libraries they will probably be syncronous and one stroke is recorded, or maybe two so a loop point will occur
- good libraries like Spitfire Audio Hans Zimmer strings have loads, and loads of articulations

So careful planning of articulations needed would be a good start before doing recording.

Just a few thoughts....

Thank you. I did not mention this in my opening post, but I actually have to create a bunch of samples that are similar to the SY77 ones, so I can't include fully sampled notes. They have to have similar loop points. For example, the brasses need to be able to be sustained indefinitely, even if that's totally unnatural. In the SY77 samples all natural decay is removed with an opposite volume compensation. You're supposed to introduce the decay through the amp envelope in the patch.

[sheaf]

Try to run the sounds through PaulXstretch (free) which will even out the changes in pitch and volume due to the stretching. You should then be able to loop the samples a lot easier while remaining the characteristics of the sound. You might want to split the transients and the sustain part of the sample, because the attack of the original sample will be smeared in time as well. I have tried this approach myself and it does work for looping, but does introduce phase issues in stereo samples.

That's an interesting idea! If I can make the attack of the instrument transition smoothly to the looped portion then this could work. The samples are all mono so I won't have any phasing issues there. Thanks!

[HAL76]

I'm looking for production techniques that might have been used by engineers in the 80s and 90s to make the samples included with rompler synths loop properly.

If you take the SY77 for example, for some samples the loop is just a single cycle waveform with wavelength that corresponds to the fundamental frequency. If you try this with any random instrument sample from the internet you'll hear the point where it starts looping can sound pretty discontinuous and jarring. Like a sudden timbre shift. They somehow managed to make this sound pretty smooth in most cases.

For other samples the loop is an actual segment of the sample, and often this too sounds really good. If you try to replicate this with a random instrument sample you find online most of the time this will introduce a significant "beating" effect as it loops. The speed of the beating then changes with the note played and this can sound really fisher price.

When creating my own loopable samples I tried to use the following principles:
- The sound is resampled with a sample rate set such that the wavelength corresponding to the fundamental frequency is an integer.
- The loop length is always a multiple of this wavelength.
- Cross fading is optionally applied to make the transition smoother (just like in Kontakt).

With just these techniques I'm not getting results that are good enough, especially with brass and woodwind samples. I tried to smooth out the harmonics with a dynamic EQ to make the overall sound more even, but this barely moves the needle.

The only technique I can think of is to implement a system described by Xavier Serra in his PhD thesis. Basically you resynthesize the sound by splitting it up into a sinusoidal and a residual part. The sinusoids for the harmonics each have an amplitude track that you can easily stabilize by smoothing it out. The residual part is resynthesized using dynamic filtering of noise. These filter coefficients can of course be time smoothed too. Hopefully this will reduce the beating effect?

I don't know how well this is going to work in practice though, and it's a lot of work to properly implement, so I'm hoping there are other techniques I can use with existing software.

Any ideas? Or did they do it simply by having excellent recordings?

They had lots of experience, but as I´ve once read in an interview on SOS with an E-Mu Engineer they frequently tried and discussed looping techniques, too.

From my perspective it´s mainly about the tool you use. Kontakt for example has a really poor loop section, while Halion works really good and automatically finds good loop points, too.

Secondly it´s a lot about experience and experimentation. Don't limit your perspectives with theories that you never prooved. Just try and experiment.

And for now, finally: have you ever sampled recordings of a choir, a string section or a synth in unison mode, for example? It's much, much easier to loop them. And I believe that duplications and subtle unison samples are also part of the sound character of ROMpler - especially since "doubling" and "layering" were an integral part of many productions at that time.

Interesting approach by the way. Let us know about the results!

[Benedict]

Two main paths:

1. Zero Crossings. Set start and end of waveforms at the same point in the duty cycle at the point where they cross the zero line (in the middle). So that could mean you set the start and end points where the Saw takes it's rise through the zero. This should be about perfect every time. Some samplers will be upset if the sample start and end coincide with the loop points so best to have a few samples either side. I often have the sample satrt way off a zero crossing so I hav DC pop for if I want it later.

2. Crossfading. Is where you set a useful start point and are more 'casual' about the end point, then let the system crossfade around those points. This can require a lot more buffer space around the loop points to use for the crossfading. If exporting that sample, you commonly have to render the results as every device can do it differently.

[slackhead]

[Shabdahbriah]

^^^ +1 SoundForge. (which I beta tested way back@v.1)

Current deal at Humble Bundle for SF 17 Pro, is well worth it (IMHO):

https://www.humblebundle.com/software/c ... e_reminder

[whyterabbyt]

I'm looking for production techniques that might have been used by engineers in the 80s and 90s to make the samples included with rompler synths loop properly.

You could always have a dig through old 80s and 90s music magazines

https://www.muzines.co.uk/mags

eg for 'looping'

https://www.muzines.co.uk/search/looping

[eLawnMust]

Well, i have made a tone of SFZ/SF2/AKAI instruments by freezing the output of VSTi in HighLife... I always sample every semitone thus nothing sounds weird or off...

HighLife places loop points automatic & does a very good job the only drawback is the limited 2-4-8 second freeze times... 2 seconds is never long enough, 4 seconds works for some, 8 seconds is best & produces the smoothest loops & most you cannot tell the looping by listening but large file sizes making a soundfont around 80 MB for a single instrument but with large HDDs now who cares...

[HAL76]

What a pleasant Tutorial. But wasn´t sampling itself the lost art? https://www.soundonsound.com/techniques ... ing-part-1

Regardless of this: looping is no longer an art today. From my perspective, it is the question of effectiveness that makes the biggest differences. Looping with an audio editor. Does that really have to be the case? And how could we maybe optimize our processes?

[audiojunkie]

Hey Sheaf,

Looping is largely a dark art. But there are some cool old school tricks that can be used to help with loops, when normal methods don’t.

1. As mentioned earlier in this thread, you can use a multitrack editor, and separate the initial attack from the loop section. Keep the attack on a different track from the loop, so that you can process it differently.

2. For the loop part, you can use tools like a compressor and a transient shaper to try to coax out a natural sounding loop.

3. If that fails, in the worst case scenario, you can bake a loop into the sample. To do this, determine where you’d like the loop to start and where you would like the loop to end. You may have to separate the initial attack and the final fade part of the sample to separate tracks. Then copy a small part of the sample, leading to the initial loop point (maybe 500 ms to 1sec in length) prior to the start loop point, and paste this to the end of your sample—overlapping the end of your planned loop. Place a linear cross fade on the overlapping part. The end result is that you have a perfect loop—baked into the loop component of your sample. Reattach your attack part of the sample, and, if needed, the fade out part of the sample. Note: it may not even be necessary to use the fade out part of the sample, since you can use the ADSR to fade the looped release.

I know that “a picture is worth a thousand words”. I used to know of a video that show how the looped release component is done. I’ll have to look and see if I can find it, and post it here.

[mrbbojahr]

Hello sheaf,

I've also been looping samples since the 90s. But it was difficult, the necessary hardware expensive, and it took a lot of trial and error. And Antares Infinity was expensive...
Everyone who worked intensively with it back then discovered their own secret recipes for creating perfect loops and preparing them in advance during processing.

Twenty years ago, I wrote a little program for myself that was supposed to automate these manual steps for me. The program still exists today (Endless WAV) and might be still useful for creating really short loops without crackling. The program has been free for many years, so if you like, you can try it out for this purpose. The program basically does what I used to do manually, step by step.
Today, the need for such short samples has almost disappeared, so hardly anyone needs such software for smallest loops anymore. And for longer loops, there are perfect autosamplers (Sample Robot, Autosampler in Logic, etc.). But if you're looking for these really short manual loops, you might want to give it a try... Of course, it's not always possible to do without crossfades, but the program can handle those too.

[beegee]

^^+2 for SoundForge.