How to make a good sampled instrument

I'm still relatively new to sampling, but I have some old synths like the Roland JX8P I'd rather like to sample before they stop working. Now, this is basically just recording notes, looping the samples and mapping them onto keys, but after a few attempts I can't seem to lose the feeling that I could save time or megabytes by being more efficient.

So... I'd thought I'd start this thread to hear what other people do. I understand there's no ultimate sampling solution and it all depends on what you want to achieve with your samples, though. I'll start out by describing what I do.

1. I select a sound on my synth. To facilitate looping, I'll disengage amplifier modulations since I can emulate that in Kontakt. Depending on the importance of the filter for the sound I'll switch that off/open it maximally.

2. I program a range of notes in my sequencer (Logic Audio) and let them play. The more 'movement' there is in a sound, the longer the notes will need to be for looping to work. I've started using 3 semi note intervals (C-D#-F#-A) per octave, but found that 6 semi-tones per octave work adequately as well).

3. I play the sequencer and record the notes as one so I can afterwards normalize it all and preserve the loudness leveles between the individual samples. This means, however, that I then need to carefully split the samples and trim lead and end silences. Is there software that helps with this?

4. Looping... this is a bit hit and miss and rather tiring. I do looping in Kontakt, but I miss built-in features like stripping parts of the sample thats not used (if I want to optimize sample use to, say, only the attack phase and the sustain loop). What software do you all use for this?

5. Mapping is rather easy in Kontakt, although I'm still getting to grips with the group-feature.

 

Hi,
Your method is remarkably similar to my own. However, I hold to the theory that loops are an economy version of long notes. Since you're sampling your own instruments, you have the option of making longer notes to begin with, hence no need to loop. Loops are economical disk space wise, if that's a problem, but if not - you asked how you could save time and this is one way. Of course, you will find many people who disagree with this method, but everyone has their own preference. To me, disk space is cheap, time is precious.
One other thing - the flex envelope is essential to making sampled pads respond like the real thing. It allows you to extend the note release much farther than the ADHSR.
That's my 2 cents worth.

 

> 2. I program a range of notes in my sequencer (Logic Audio)
> and let them play. The more 'movement' there is in a sound,
> the longer the notes will need to be for looping to work. I've
> started using 3 semi note intervals (C-D#-F#-A) per octave, but
> found that 6 semi-tones per octave work adequately as well).

This really depends on the sound. I would suggest that 3-6 semitones per octave is probably too much when sampling a synthesizer. (When sampling an acoustic instrument, it can be too little, because acoustic instruments vary significantly based on their range and also their velocity.)

A synthesizer is basically just speeding the waveform up and down to create different pitches. Kontakt can do that easily. Of course this is an oversimplification because filters and effects and routings can play a huge part. But 90% of the time, I've seen people sample synthesizers TOO detailed, resulting in a bloated patch that would sound just as good if they sampled middle C and let Kontakt stretch it.

> 3. I play the sequencer and record the notes as one so I can
> afterwards normalize it all and preserve the loudness leveles
> between the individual samples. This means, however, that I
> then need to carefully split the samples and trim lead and end
> silences. Is there software that helps with this?

A few notes here. First of all, before recording, play the MIDI track and watch your audio input level in the sequencer. You want to make sure the very loudest notes get as close as possible to 0dB without actually clipping. It is vital you do this, otherwise the normalization process (described next) will magnify whatever noise there was, resulting in a noisier patch.

Second, record the audio in. Then, normalize the ENTIRE audio track. I would do this before splitting it up into individual notes. I would NOT normalize individual notes. Some notes (or a range of the keyboard) will naturally be slightly louder or softer than another. Normalizing individual notes makes everything exactly the same, taking away from the distinctiveness of the patch. Normalizing the full track of audio just makes sure the whole thing is as loud as possible.

> 4. Looping... this is a bit hit and miss and rather tiring. I do
> looping in Kontakt, but I miss built-in features like stripping
> parts of the sample thats not used (if I want to optimize sample
> use to, say, only the attack phase and the sustain loop). What
> software do you all use for this?

I agree, Kontakt is OK but obviously not a full-featured sample editor. I use Peak, and have heard great things about Spark. You could probably find free and shareware tools that would do an OK job too (better than Kontakt).

One thing that's very very important to recommend here is that, on every sample, you zoom in to the attack and crop the sample so it starts right where sound begins. This is the #1 mistake I see with amateur sample designers. Not cropping the sample leaves silence before the attack of the note, which users will later think is latency and sluggishness. Crop them tight!

That does it for my general observations. Some personal opinions to add to this: the world doesn't really need clean samples of an old synthesizer in the sense of "here's patch #1, here's patch #2, woo hoo, here's my entire DX7 ported into Kontakt!" because (1) these exist already, (2) better versions exist in software, such as NI's own FM7 or Reaktor. (I have a killer Reaktor-programmed emulation of a Roland JP-8000! See www.dashsynthesis.com for details on that.) So I would discourage users from simply sampling patches off a hardware keyboard. It's not that useful, unless it's a really unique patch or a really awesome keyboard (like a Rhodes or a Wurly or something electro-acoustic).

There is a little more merit in sampling unique patches that you've created in a keyboard. That's a little more interesting. A step above that would be to actually combine patches from various synthesizers to make one huge mondo patch. I've done, for example, humungous pads by layering 3 pads from my Roland keyboard, plus 3 from a Korg, 3 from a Yamaha, 3 from an Alesis...etc...thus I'm coming up with combinations that no one else will probably ever think of or be able to replicate. That's more useful than just taking my Roland and saying, "You know, I'd like to sample that Roland piano in bank A-001..." which may be a useful exercise in sampling, but in the end it's just faster to play it from the Roland.

Lots of opinions here, hope some of it was helpful.

 

One thing I forgot to mention in my above essay...

When I was talking about the intervals at which you should sample synth sounds, there's nothing wrong with recording lots of intervals (3-4 per octave) and there is extra work involved in preparing the samples, but I would always try your final Kontakt patch with fewer and fewer samples until you can noticeably tell that you're skimping on samples.

There's no point in sampling every single half-step to create a patch that's dozens of megabytes in size, if you could have gotten away with the same patch with one sample per octave, and left Kontakt to do the stretching. (Always put the key note in the middle and allow Kontakt to stretch both up and down.)

In my experience, when sampling synths, unless it's a very complicated envelope like something that would come out of Absynth, I don't need more than 1-3 samples per octave, sometimes even less than that.

But experiment. Oversample at first, then prune samples you don't need.

 

Thanks, some good tips there, as well as good to hear that I'm doing some thing ok already...

Pdgood - looping is quite necessary in my view, at least when you're playing instruments that don't fade out naturally, like a piano. If you're playing a sampled instrument like a synth pad, lead or an organ for that matter, I'd hate to have the sound drop away holding a note for longer than I anticipated. Having said that, I can imagine some sounds where that could be a solution. I'm also going to check out the flex envelope some more.

Ddas - good points about the amount of samples needed per octave. Unfortunately, I started off with my JX8P playing a sound where the two oscillators cross-modulated and then were sent through the internal chorus effect, making them sound pretty different from note to note and having loop-times of up to seven second. Having said that, I got by with just the C and the F# per octave, so I'll certainly watch how much I need.
The reason I sample my synth is quite simple - I like the sound, it's old and I'd like to have an alternative when it doesn't work anymore, and also - I can play multitimbral, with a higher polyphony and using Kontakt's effects and filters.