Timestretching and SRC In Vintage Samplers And Modern Equipment / Software

I’m looking into the tuning samples and sample rate conversion thing. Based on the principle where people feed pitched-up samples into their sp-1200’s or what have you, and pitch them down in the sampler to get a richer, strange sound.

Or into the 90’s rackmount samplers and play the tone over the keyboard, to get the specific sound of that sampler.

I’ll be doing some sound tests today with the samplers I have and the pc / ipad software I own. To see which has the most pleasing effect. But I also want to understand better, the technical side of what makes the 80’s sampler sound like it does and the 90’s sampler.

What makes a daw like ableton have a very clean effect when tuning a sample as oppose to the vintage samplers?

Is it based on the chip, or also on the amps or other electrical components of the vintage samplers?

Am I right in thinking most ‘bitcrush’ effects don’t do actual sample rate conversion, but mimic the effects of it?

Are there some recommended plugins that could do the same job, in a real way, as a vintage sampler?

thanks, I’m really curious

7 Likes

I am not an expert on a topic but I think few things might give older samplers their specific sound from the digital side:

  • lower sample rate - the lower the sample rate the narrower the band of frequencies that can be reproduced by sampler without introducing aliasing. For example sample rate of 44,1kHz can “reproduce” frequencies up to 22050Hz but if sampler has sample rate of 22,05kHz then it can only reproduce up to 11025Hz so basically sounds will either sound low passed or frequencies that were not in original signal will be introduced through aliasing so it definitely changes the “characteristics” of sampled sound
  • lower bit depth - bitrate determines signal/noise ratio and general possible dynamics of sound so the lower the bitrate the more compressed it can sound
  • General algorithms for pitching sounds up/down and how they change the sampled sound. For example when you pitch sound down you basically need to calculate values that are not stored (because the slower you playback the sound the more “values” you need to have per cycle) so the algorithm to “interpolate” these values can also introduce stuff that were not present in original signal. Similiar thing can also happen when you pitch sounds up.
    Also some samplers were repitching the sample by changing its internal sample rate (if I recall correctly the Fairlight was doing it that way).

Generally it is a very broad and fascinating topic so hopefully someone could shed some more light :smiley:
EDIT: From the quick look it seems that SOS had an interesting series of articles about sampling with a lot of technicalities included https://www.soundonsound.com/techniques/lost-art-sampling-part-1

5 Likes

Cool! Always have been interested in this topic, especially with regards to tape machine pitching, and how that alters the formant in such a unique way. Valhalla has an interesting article.

For hardware samplers, I assumed it was the result of the sample rate / bit depth of the A>D. Certainly other components surrounding the converter will impart something on the sound as well. I’ve always been enamored by early digital pitch shifting - particularly the Eventide 949. I have nothing else that sounds quite like it.

I love the pitched down sound of the op1 4 track, but found that it exposes a lot of high frequency artifacts. It also changes the formant in a much different way than my cassette 4 track. Surely this is a result of many things, but I would assume the A>D has a big influence on the sound.

Following!

2 Likes

Yeah just to add to your third point, when you sample at a higher pitch than you want to end up with and pitch down (by say, an octave) on a sampler you’re effectively reducing the samplerate of the initial sample. If you’re halving the pitch of the sample you’re also halving the sample rate vs having sampled at that pitch in the first place. People used to do this to save on sample memory when you might have only had a few MB to play with.

4 Likes

I’ve been wondering about this a lot lately too! I’m especially fond of the op-1 slow-tape sound. Also I’ve been wondering more about what’s happening under the hood of modern samplers such as in the Assimil8or, the Octatrack and especially Softcut. I think a lot of the sounds that spark my ears in modern music are the results of sample mangling more so than traditional synthesis. Very keen to hear the experts weigh in on this.

1 Like

I’ve been extremely interested in understanding this process from a computer science standpoint.

as others have hinted, you’re essentially reproducing a digital to analog conversion process at a lower sampling rate.

but it’s not that simple. there’s way more nuance in samplers. If you’ve ever used redux in ableton or degrade~ in max you’ll know that you can’t get anywhere close to the sound with straight sample rate reduction. there’s more nuance that happens in other parts of the conversion process, maybe on the analog side as well. I’ve just started scratching the surface, but experimenting with interpolation has been very rewarding.

and yes softcut performs this process extremely well (thought it doesn’t emulate anything in particular). I’d highly recommend looking at the source code if this is something you want to understand. From my understanding it’s pretty much just sample rate conversion, interpolation, some soft-clipping, and an input/antialiasing filter that give it the sampling sound it has.

the vulf compressor was built around careful studies of the SP-303. been wanting to try it for a while.

4 Likes

Thanks all, I’ll have some reading up to do.

I got triggered on learning about the technical side of this by seeing these two videos by Junkie XL on samplers:

I’ve been broadly obsessed with this stuff for as long as I can remember, having basically grown up on music that was the result of vintage samplers.

The best approximation of the sounds I love that I’ve found is in D16’s Decimort 2 plugin

On the surface, the sound of the sampler comes from a few factors - the main 2 being:

  • maximum sample rate
    and
  • resolution (or bit depth)

In hip-hop terms, I spent most of teens hankering for an SP1200 as those were the first samplers I was really aware of. In my mind there was something golden about 12-bit at 26khz (max) and it didn’t matter that my PC was capable of far higher quality, it didn’t sound “right” to my ears. I even tried saving my sounds in 12-bit (mono, of course) but they just sounded a bit crap - definitely not “golden” like the sacred sounds of the SP.

I wouldn’t pretend to be any kind of authority on the subject at all - just a very enthusiastic sampler nerd - but I do know that a whole variety of things also contributed to the sound of those hardware units, including things such as: specific chipsets; AD converters and even filters in some cases

The point made earlier about sampling things at a higher speed and then pitching the samples back down as a method of getting around small sample times is an excellent and very important point. Sample something at the speed you want to use it at will produce a sound which has a noticeably different quality to sampling an LP at 45rpm +8 (for example).

One thing I never considered doing at the time was something J-Zone once mentioned: sampling from tape using high speed dubbing! That’s genius right there!

2 Likes

One thing that I discovered in my recent venture into sampling is that some early samplers (I think the Ensoniq EPS, but can’t remember definitively) do not change the sample rate to re-pitch the sample. From what I read, in this case, they are adding in information (redundant samples) to reduce the pitch. Someone who has studied the architecture of these more directly could explain it better than I am, but I do know from listening that any time I have used a sampler that used this method it had a distinct sound, and could be programmed differently. On my ensoniq mirage, you really want some kind of keyboard tracking on the filter to soften up the aliasing (or not!) so it will have a very different sound than the EPS (granted that would be true regardless, but I mean that it has a very different morphology).

2 Likes

few random thoughts

plenty of artifacts from both algorithms and hardware components at both capture and playback. the design of ADCs/DACs themselves have changed a lot.

since the 80’s most low-cost ADCs are sigma-delta architecture. in a very small nutshell, this means starting with a fast 1-bit sampler (a comparator), applying lowpass filtering, doing this a number of times, and performing error correction. this basic architecture has been drastically improved many times in the last 30 years (e.g. much better decimation filters: more gates on a die -> more taps on a FIR.) Analog Devices website has some deep-dive white papers hosted that are hidden but searchable if you are interested.

many older samplers had additional analog filtering stages, sometimes with interesting and arbitrary uses, like linn lm-1 letting through some grit (8-bit samples) in the attack portion of envelope.

ableton has a variety of algos to apply. resampling for pitch change plus time change can be done without significant degradation using windowed-sinc interpolation, and on modern computers this is fast enough to seem realtime (but there is a fundamental trade off between window size and quality and window size implies latency.) “warping” usually means doing something more exotic using frequency-domain representations, time-domain granulation, or a combination. (that’s a large topic in itself but not really related.)

Softcut

uses 3rd-order resampling by default (though i keep meaning to add runtime switches for lower orders.) this is a common compromise for realtime resampling, it has less distortion than linear and is nearly as fast to compute.

(softcut is very straightforward; if it has any interesting features they are about record+ partial erase behavior during cross fades, which is surprisingly hard to do in a natural-sounding way without expensive input analysis. but it’s main reason for existence is just to encapsulate some common stuff that is fiddly to get right in high-level DSP environments.)

usually that means doing something with the sample depth, not sample rate. sometimes it is simple rounding, sometimes an attempt to emulate old hardware by using a nonlinear quantization like u-law encoding. like old telephones, old samplers may have also applied dynamic range compression at capture stage to make the most of limited bit depth with these encodings.

if you are interested in specific spectral effects of quantization, sampling &c, i would look at a good modern DSP book. zolzer wrote a good one that is very clear and to-the-point, in particular chapter 3 for this stuff. if you are interested in the math behind it, hamming’s “numerical methods” is the classic reference.

that sounds like resampling. if the new, intermediate values are simply repeated from existing values, that’s zero-order inteprolation. if they are extrapolated in some way, that’s some other kind of interpolation.

14 Likes
6 Likes

i can’t add much technical info to this conversation… but for anyone with an er-301 you can get nice results by setting the sample player interpolation to “none” and inserting a fast-clocked (20khz) sample and hold unit after your sound source.

1 Like

As others have mentioned, A>D, bit and sample rate, and D>A as well as the filters…
Suspect accurate filter tracking is crucial when pitching down

Presume you are aware of TAL Sampler plugin?
They have a block diagram of eg how they emulate the EMU II DAC:
https://tal-software.com/products/tal-sampler

  • Vintage DAC modes (Emu II, AM6070, S1000, Sample Hold, Linear, Clean).
4 Likes

That was a REALLY satisfying read. Thanks

Sorry I’m not more technically versed, and this is a bit off topic since it’s analog but this seems relevant to the downsampling convo and was interesting to me when i saw it and maybe will provide some inspiration:

3 Likes

For interest. Here’s a sound test of the Korg Volca Sample, the Zoom Sampletrak ST-224 and Ableton Live. Converting a precussive loop down twelve semitones. So a modern hardware sampler, a late nineties one and the daw:

2 Likes

Kleggrand has Degrader, which has become one of my favorite downsampler/emulator type plug in. Super fun!

3 Likes