Your Sampling strategies for keyboard instruments?

[SWAN808]

Im curious about how people prefer to sample their keyboards (MIDI hardware)...instruments like pianos, rhodes, and synths...ROMplers for their own use...or synths...

I have been testing and I have to be honest and say I generally prefer sampling per note rather than stretching...I know some may not agree with me but in a perfect world I would sample several round robins per note, and several velocities...

However - that makes for rather big libraries...

Has anyone done tests and felt stretching is fine? Or are you happy with just making a bigger library because hard disk space is cheap...

[hollowsun]

Every minor third can be an effective way of representing an instrument well without compromising it and cuts library size down dramatically (2/3rds). For example, sample on C, D#, F#, A of every octave spanned B-C#, D-E, F-G, G#-A#. That way, nothing is ever transposed more than a semitone either way

An advance on that is to sample chromatically for the main playing area of the instrument in question but every minor third for those notes that won't be played that often. So, for example, chromatically for C2-C4 but less either side of those ranges.

Round robins - can be effective but you can also use velocity > sample start, velocity > filter cutoff, etc.. All sorts of 'old skool' trick you can use to get the size of your library down (this from someone who squeezed grand pianos onto 1.4Mb floppy disks!).

And also worth bearing in mind how the sound will be used. A piano in glorious isolation might require more work (*) but something that will just be in there in the mix might not need such attention to detail.

And there's bit depth and sample rate. Believe it or not, there's no law that says you have to record at at 48 or 96kHz/24-bit. In fact, for many instruments, that's just a waste of disk space and CPU usage - e.g. why the hell record a kick drum at 96kHz when there's bugger all above 4kHz (if that ... at best)?

Unfortunately, many judge the quality of a library by its size and its spec numbers rather than how good it actually sounds. I've played 96/24 stuff that sounded like like shite (but looks great on paper) and 44/16 stuff (or less) that sounds fabulous.

(*) Actually, making compact, well optimised sampled instruments can be more work than just recording GBs of samples and lobbing them into a sampler.

HTH

Cheers,


Stephen

[Sasja]

Stephen, sometimes 96kHz/24-bit is demanded for HD audio in movies. For regular audio it isn't a big of a deal I guess, it cuts the size nearly in half when doing it at 16-bit.

[CriticalVibrations]

I've been working on sampling my QS8...personally, I like to record at 48/24 on all 88 keys. Hard drive space is cheap, but 96kHz just seems like overkill for my own purposes.

But, I noticed that a lot of the factory presets aren't accurately distributed across the board, especially at the high end. For example, the right-most octave of the keyboard might be a duplicate of the next one down, and not an octave higher as should be. So, when I find those discrepancies, I just chuck those samples.

[SWAN808]

Hi thanks for the replies especially Mr Hollowsun. Interesting - with the piano - I take it you simulated a more muted velocity hit with the filter cutoff? But sample start - is that for a similar thing? Can this be done simply in Kontakt-or is it a complicated script?

I think for a sample library - if it is recorded at a good level - is 16 bit good enough? Do you need the extra dynamic range if you only do a few velocity layers? That seems like a reasonable way to save space...every minor third - I might test that against every key and see if I can tell...

I like the idea of sampling per key in important areas.

Whilst you are here I wonder if I can ask a question about your excellent YouKnow sample synth. I noticed on playback the start point jumping around - so I assume its random start on the sample - to bring in some analogue style variation. However the sample setting of kontakt is DFD - which I thought was not possible with random start? Can I also ask - why sample so long on some of those waves?

[hollowsun]

Stephen, sometimes 96kHz/24-bit is demanded for HD audio in movies.

Yes ... and we're all working in that field, aren't we?!

For all practical purposes, 44/16 is more than enough for 'samples' even if the user is working at 96/24 - the DAW will (or should) take care of it all. Many of my customers work in this area without complaint and there's an awful lot of "Emperor's New Clothes" about this all this silly specmanship.

There's a case for 'recordings' at 48/24 but not (so much) for 'samples'.

Cheers,


Stephen

[niktu]

Yes, interesting point. I hear you on the 'specmanship' thing. And I agree about 16/44 samples too. I've been bowled over by the quality of some 16bit libraries. I love the Hollowsun M1000 library and Phaedra is another work of art. I just get lost in the sound-design and the qualities as a synth rather than get my knickers in a knot because they're not 24-bit samples.

I'm sampling some analogs at the moment and just gave up on the idea of doing higher sampling rates. The poor things generally struggle to reproduce much over 12k anyway (well, most of the time).

However, I've spent a lot of time laboring mentally over the '48 v 44' issue. I consider that 44.1 is only tied to CD - of which, is a dying format. I'm sure 44.1 will hang around for a while after that, thanks to the sample-library market but aside from that, I can't see much point working in 44.1 when CD isn't the end-format.

As such, I've decided to do my stuff at 24b/48k. I might be very wrong but I just can't see 44.1 lasting into the future. 48k just seems a better choice for the long term to me, and is permenently buoyed by the film & tv industry. I think musicians will follow suit eventually, when CD fades away altogether. Crystal ball anyone?

With that said, it probably doesn't matter anyway (48 v 44). As Stephen said, the DAW will take care of differing sample rates & bit depths so in reality, 24/48, 24/44 & 16/44 can all live together in the same project without issue. We can have it all.

BTW 16bit will save you some substantial hard drive space but 24Bit will give you more headroom (quieter noise floor) - if you need it. Good luck.

[SWAN808]

niktu - how are you sampling your analogues - do you use round robin? Are you stretching samples?

[mkdr]

My M1 sounds better than most of the GIGAbyte soundlibraries. It's used on many many records, movies and tv stuff. And it's not 24/96. Just tells that good design is more important than megabytes.

The size doesn't matter!

[niktu]

I'm doing 3 round-robins (though I sample 5+ for problem takes & random artefacts), 6 octaves (or there abouts), minor third intervals and aiming my peak levels at around -3dBFS. No normalising, no stretching and generally avoiding post-processing as much as possible.

I'm keeping the minor pitch variations (nuance?) between notes on my DCO synths and going to try Expert Sleepers Silent Way when I have another crack at my VCO mono synths. It apparently has some clever 'listening' trick to send out the correct pitch. Another KVR poster suggested it in another thread and have since looked into it and it sounds promising. Looking forward to trying it out.

mkdr - Indeed.

[hollowsun]

Hi thanks for the replies especially Mr Hollowsun. Interesting - with the piano - I take it you simulated a more muted velocity hit with the filter cutoff? But sample start - is that for a similar thing? Can this be done simply in Kontakt-or is it a complicated script?

It can be done as a standard modulation function without scripting.

I think for a sample library - if it is recorded at a good level - is 16 bit good enough?

IMO, yes for the most part.

Whilst you are here I wonder if I can ask a question about your excellent YouKnow sample synth. I noticed on playback the start point jumping around - so I assume its random start on the sample - to bring in some analogue style variation.

Correct

However the sample setting of kontakt is DFD - which I thought was not possible with random start?

Not sure about that. Maybe Evildragon/Mario might drop in to make comments - he scripts my stuff.

Can I also ask - why sample so long on some of those waves?

To give them time to breathe and develop and create a good and generous loop. Back in the early days with limited memory/storage, I'd make a sound and it was bloody hard work to loop, for example, sampling 'Bong'. Inevitably, it would end up as 'Bongongngongong' or 'Bongngngng' or even 'Bonnnnnnnnn', whatever. F'k'n PITA., With such generous memory available today, it's good to be able to let a sound breathe. And with the slightly random and asynchronous nature of the famous Roland chorus (not just the Juno but the VP330 as well) AND stereo, you need that time to let the sound 'happen'.

A simple sine wave I could capture and edit to a few Kbs or whatever of a single cycle. But define "simple sine wave"! A digitally generated sine, easy ... but a sine wave from some old valve/tube tone generator (which I normally deal with), I give them some space to - as I say - breathe because they have a kind of life of their own. Sounds daft, I know but it's how it is.

It all depends on the source material and what it's it's intended for.

All IMO, of course.

Cheers,


Stephen

[Sasja]

Personally I always record at 96kHz/24-bit, then I burn it to a disc, then down-sample everything for creating an instrument. Because maybe later on I might decide to release a 96kHz/24-bit version of it, or maybe I need it in some way. I can't just re-record everything.

I don't really care much about 24/16 bit or 96/48/44kHz. But I must keep in mind that some people are "specs readers" and demand the "best", especially in the movie industry, foley artists etc. And who knows, 5 years from now the source audio *must* be in HD 96kHz/24-bit. Yeah you can up-sample it, but I won't take that risk.

[SWAN808]

I'm doing 3 round-robins (though I sample 5+ for problem takes & random artefacts), 6 octaves (or there abouts), minor third intervals and aiming my peak levels at around -3dBFS. No normalising, no stretching and generally avoiding post-processing as much as possible.

I'm keeping the minor pitch variations (nuance?) between notes on my DCO synths and going to try Expert Sleepers Silent Way when I have another crack at my VCO mono synths. It apparently has some clever 'listening' trick to send out the correct pitch. Another KVR poster suggested it in another thread and have since looked into it and it sounds promising. Looking forward to trying it out.

mkdr - Indeed.

thats interesting - doing round robins but not sample per key...I might try that...how long do you sample each note for?

Personally I always record at 96kHz/24-bit, then I burn it to a disc, then down-sample everything for creating an instrument. Because maybe later on I might decide to release a 96kHz/24-bit version of it, or maybe I need it in some way. I can't just re-record everything.

I don't really care much about 24/16 bit or 96/48/44kHz. But I must keep in mind that some people are "specs readers" and demand the "best", especially in the movie industry, foley artists etc. And who knows, 5 years from now the source audio *must* be in HD 96kHz/24-bit. Yeah you can up-sample it, but I won't take that risk.

thats a cool future proofing strategy - what do you use to downsample? Is there any potential quality difference with downsampling vs sampling at 44.1?

To give them time to breathe and develop and create a good and generous loop. Back in the early days with limited memory/storage, I'd make a sound and it was bloody hard work to loop, for example, sampling 'Bong'. Inevitably, it would end up as 'Bongongngongong' or 'Bongngngng' or even 'Bonnnnnnnnn', whatever. F'k'n PITA., With such generous memory available today, it's good to be able to let a sound breathe. And with the slightly random and asynchronous nature of the famous Roland chorus (not just the Juno but the VP330 as well) AND stereo, you need that time to let the sound 'happen'.

Hollowsun - thanks thats interesting stuff. One thing I found using the YouKnow was I liked the random start point - giving some variation. I thought this was a very clever way and maybe I could avoid Round Robin. However with YouKnow I found if you have a longer release - and hit the same note - there can be some phasing because of the same sample overlaying...? I wonder if there is a way to avoid this...

[niktu]

thats interesting - doing round robins but not sample per key...I might try that...how long do you sample each note for?

My oscillator samples are generally around 4 to 4.5 sec in length but I have sampled patches 30-45 secs long in some cases. It just depends on the patch.

Regarding RRs v every note sampled, my thinking for RRs is that repetitive chords and single-note lines will have some slight variation to them. Sampling every note, doesn't quite have the same appeal and seems a tad unnecessary to me. Though some would argue the same could be said for RRs too (unnecessary).

I'm doing round robins for now but if I think a synth instrument doesn't benefit enough from it, I'll just create a version with single-round samples and maybe try the random-start approach as discussed earlier.

I would luv to hear some other perspectives on RRs and the every-note-sampled approach for keyboard/synth instruments.

[hollowsun]

I wonder if there is a way to avoid this...

There's not really (well, other than not using long releases ). I've had it with 'proper' polysynths and sampled synth (and other) sets. Round Robin might/could go some way to overcoming this but can introduce other artefacts!

Cheers,


Stephen