I sampled a piano. Which software editor/tool for automatic editing? Free/paid?

Sampler and Sampling discussion (techniques, tips and tricks, etc.)
User avatar
fmr
KVRAF
10338 posts since 16 Mar, 2003 from Porto - Portugal

Post Thu Jul 22, 2021 9:51 am

enroe wrote:
Thu Jul 22, 2021 9:11 am
fmr wrote:
Wed Jul 21, 2021 8:31 am
All the samplers I worked with that "supported" SFZ only read the sample regions.

The only free sampler I saw in the last years that's free and worth something is
TX16Wx. All the rest is basically trash.
Well, with this you are impressively demonstrating that you have
no idea - and that you are still verbose here. :?
For me this is a huge contradiction - and that puts you on my
ignore list. :wink:
Whatever man... You already demonstrated with your "lists" that you have no idea what a real sampler is. That's why you keep talking about SFZ. :roll:

I just posted because the OP is clearly a noob, and with all your biased statements you risk to cause him a lot of pain and wasted work.
Fernando (FMR)

AnX
KVRAF
9841 posts since 17 Nov, 2015

Post Thu Jul 22, 2021 10:23 am

fmr wrote:
Wed Jul 21, 2021 1:25 am
instrument. Things like envelopes, filters, modulations, etc. have to be defined FOR EACH SAMPLER
isnt that the job of the Sampler, not the samples? you load a sample set and set your own envs, filters, mods etc on the Sampler gui....?

just interejecting, not taking a side :wink:

User avatar
fmr
KVRAF
10338 posts since 16 Mar, 2003 from Porto - Portugal

Post Thu Jul 22, 2021 11:01 am

AnX wrote:
Thu Jul 22, 2021 10:23 am
fmr wrote:
Wed Jul 21, 2021 1:25 am
instrument. Things like envelopes, filters, modulations, etc. have to be defined FOR EACH SAMPLER
isnt that the job of the Sampler, not the samples? you load a sample set and set your own envs, filters, mods etc on the Sampler gui....?

just interejecting, not taking a side :wink:
Yes, that's the job of sampler. But that's my point, exactly.

You can't create anything meaningful with just a sample set (I mean, you can, sort of, but it will be raw and very basic, totaly dependent of the samples). To create really good quality sample based instruments, you have to rely deeply on the sampler engine (things like transposing algorithms, fiter types, modulation rountings, modulation matrices, etc. That's why it is stupid to say that you can create anything worth mentioning with a generic format - you simply can't.

You can create sample maps in SFZ, and these can be useful as a foundation for creating sample based instruments. But they are just like the naked oscillators (saw, square, pulse) of a synth. Everything else has to be programmed on the sampler itself, and each sampler is a different case.
Fernando (FMR)

AnX
KVRAF
9841 posts since 17 Nov, 2015

Post Thu Jul 22, 2021 11:16 am

yes :tu:

(aside from vel articulations/round Robins etc)

KVRist
37 posts since 5 Mar, 2021

Post Thu Jul 22, 2021 5:13 pm

fmr wrote:
Thu Jul 22, 2021 7:28 am
I did that because I already had the keymaps. Not that I was forced to. TAL Sampler allows you to create your keymaps (as does any DECENT sampler). As a matter of fact, sometimes I changed the keymaps after importing them. This is the key word here: ADAPT. Each sampler has its own peculiarities, which demand you to sometimes adapt the sample maps (and pretty much everything).

Using a text editor to program patches? Really? And how do you check the values for filter, envelopes, LFOs, etc? You type, load, play, go back, type again, load again, play again? :lol:

And are you the one that say it would be useless to change computer to save a couple hundred dollars. :dog:
As much as I like TAL Sampler, its key map functions are pretty clunky. For any instrument using more than a couple of samples, I map it using a basic temporary SFZ and import it. You know, I ADAPT.

And yes, as alien as it may seem to you, we make patches by typing. It works very well, is very versatile, and often it's faster than a graphical UI.
Again, it's not for everybody. But it's really not that strange, Kontakt programming is mostly KSP scripting these days, and much more cumbersome. Certainly not lol-worthy.

See, these kind responses is what make you appear childish and naive, even though you might have done this kind of thing for ages.
I appreciate that you brought in the Hollow Sun patches for TAL Sampler, but the patch programming for those is spotty at best. Response to velocity, mod and pitch wheels is essentially non existing, and filter, amp, LFOs, FX and ADC/DAC settings aren't particularly well thought through, if at all. Many of the regions in a lot of the mapping were overlapping, and you know that TAL Sampler doesn't support that.
I've spent many hours reprogramming the whole set of raw samples with better patch data.

If I had access to the WAV samples for the Hollow Sun set (and not just the TAL format ones), I'd love to take on the challenge of remaking the library as SFZ (to be used in a full-featured player like Sfizz), and see which patches are better, yours on TAL Sampler, or mine on Sfizz.

I'm not saying this to be a jerk. I'm just asking you to get off your high horse and exercise some humility.

User avatar
fmr
KVRAF
10338 posts since 16 Mar, 2003 from Porto - Portugal

Post Fri Jul 23, 2021 2:26 am

MTorn wrote:
Thu Jul 22, 2021 5:13 pm
I appreciate that you brought in the Hollow Sun patches for TAL Sampler, but the patch programming for those is spotty at best. Response to velocity, mod and pitch wheels is essentially non existing, and filter, amp, LFOs, FX and ADC/DAC settings aren't particularly well thought through, if at all. Many of the regions in a lot of the mapping were overlapping, and you know that TAL Sampler doesn't support that.
I've spent many hours reprogramming the whole set of raw samples with better patch data.
Wow... You want to make this personal? :roll: Please be my guest, and make those available. It's the minimum you can do, after what you wrote because, you know, I made those available FOR FREE, and it would be good if someone who can come with something better out of them shares the work also.

Not that I feel I have to excuse myself, but since you brought the subject to the table, when I programmed those TAL Sampler was still in beta. Several current features were even non-existent. For example, I chose the E-Mu DAC because the S1000 was simply not there at the time. If it was, I would have chosen that, since the samples came from the S1000. Response to velocity, mod and pitch wheels were most of the time intentionally left out because, once again, they depend a lot on the keyboard controller in use, and I am used to have to reprogram those in basically everything. Regarding sample maps with overlapping regions - During beta, it was under discussion the support of overlapping regions. For obvious reasons, I was defending that, and I prepared the sample maps for that possibility (and also because I wanted to use them in other samplers, of course). Overlap ended up not being supported, but since it didn't affect playback, I saw no reason to change all the sample maps.

Envelopes, LFOs, modulation behavior, filters behavior, etc. have been changing slightly from version to version. The same happened with the TAL U-NO-LX. Many presets I programmed on version 1 had to be recalibrated for the most current version. That's life. I also have to do that for sample libraries I bought for Kontakt, every now and then. But it's OK. If you perfected them, feel free to make them available, I will be the first one to thank you.

I know that there are certainly a lot of things that could be perfected. If I was going through that, I would certainly change a lot of them, and make them better. I am the first one to criticize my work, and I am certainly humble enough to know that it's far from perfect. I even have been perfecting it for my own use through time.

See... I am humble enough to accept your critics. OTOH you are arrogant enough to say that you could come with something better out of that "Sfizzz" thing. So do it - sample the naked output of TAL Sampler for a couple of them, if you aren't knowledgeable enough to convert the sample format ;-) (if you are someone able to create Kontakt scripts, you should, but anyway), create your own samples, and create your own sample maps. For obvious reasons, I can't share those samples, and you know it.

Funny thing - with all these you simply proved my point. Coming up with something in SFZ is basically pointless and useless, because the end user will have to program the instruments FROM SCRATCH in the sampler of choice :lol:

Unless you simply want to share THE SAMPLES: For that it may be a valid choice.
Last edited by fmr on Fri Jul 23, 2021 10:11 am, edited 1 time in total.
Fernando (FMR)

KVRAF
1584 posts since 19 Mar, 2008 from germany

Post Fri Jul 23, 2021 9:03 am

About one user here:
MTorn wrote:
Thu Jul 22, 2021 5:13 pm
See, these kind responses is what make you appear childish and naive, even though you might have done this kind of thing for ages.
Your tone is very diplomatic. I appreciate that! :clap:

Nevertheless, it is important that reading newcomers and
newbies recognize the value of statements made by certain
contributors.

So if statements are obviously wrong - also recognizable
by crude generalizations, defamation and obvious contradictions
- then that is extremely destructive - and then you have to
clearly state such things. 8)

And I also recommend not preparing the field for such trolls.
This works by putting them on your ignore list (user control
panel -> friends & foes -> manage foes) and basically not
reacting to their posts. :tu:
free mp3s + info: andy-enroe.de songs + weird stuff: enroe.de

KVRist
37 posts since 20 Sep, 2010 from Paris

Post Wed Jul 28, 2021 6:33 am

I did the same work for my organs (B3, Vox Continental, Farfisa)
and the same way.
I used Adobe Audition to cut accurately.
In addition, it allowed me to record a noise fingerprint in an internote part and then remove the noise without deterioration.
To put it all together, I used Kontakt ... the most practical.
The advantage of Kontakt is that I was able to convert it with Awave Studio for my Motif XS.

KVRist
145 posts since 5 May, 2020

Post Wed Jul 28, 2021 11:37 am

Back to the original question.

96 KHz is overkill. Almost everyone uses either 44.1 or 48. Cut the file size in half and nobody will notice the difference. Consider using free r8brain to convert the rate; Audacity can do it too.

15 velocity layers is more than I ever use, but whatever works for you. I find that real pianos vary a lot more than digital pianos with velocity. I assume you recorded a digital piano, not miking a Yamaha.

I wrote python scripts to do #1 and #2 and use them a lot. I sample my favorite patches on my digital keyboards so that I can do recording without having to render MIDI to audio with the keyboard attached. This makes it a lot easier to switch between keyboards and guitars when recording, and rendering is faster after a MIDI edit, etc. My script expects each velocity layer to be in a separate file, so if you recorded the same note 15 times and then moved on to the next, it won't help. But it trims each note tight, determines the pitch, and names the file. The second script builds the sfz file based on the file names, automatically mapping the zones, with a little input from a text control file (telling it how loud each layer is.) You'd have to install Python which sadly isn't trivial, and it's definitely nerdware, but if you don't find a solution you're happy with post back and I'll point you to it. Oh, my script also assumes each layer file is normalized, which is generally a good idea because it makes mapping the volume contours a lot easier -- it's how SFZ normally works.

Looping is the most time consuming part, so it's best to have full-length samples. At 12 seconds you'd have to loop the loud low notes, or else fade them out (which my script does not do, but it could be updated to do that.) Or just don't end a song with a big chord that fades out. It's rare in a real performance to have a note that lasts longer than 12 seconds. Yet when I've had notes the cut out after 10 due to a tool I used, it'd bother me when I'm just noodling. Practically speaking, you can ignore the tails at first, and come back and fix them later if they bug you. (I just pulled those samples into Audacity and faded out the last 4 seconds. This is less than ideal for a number of reasons, but never matters during a real performance, and avoids bugging me. It's not something I'd do for a sample set for distribution but it's fine for my own use. I've since corrected that problem at the source, but haven't bothered to re-do the earlier sample sets.)

I find that if a piano is recorded well, especially with lots of layers, you don't need to fiddle with filters. You want the result to sound like the original! With fewer layers, we use filters to adjust the timbre based on velocity. (I find that a fairly simple filter applied equally to all notes works remarkably well for most digital pianos, with one velocity layer. Partly that's due to the nature of most digital pianos, which are very "regular" and lack the peculiarities of many real pianos. Regardless I generally sample 7 layers when sampling my keyboards for my own use.)

Instruments like piano are a lot easier to sample than instruments like violin. Due to the simple attack-decay-release profile, you dodge a lot of issues that require advanced capabilities in a sample player. SFZ works just great and is easy to use for instruments like pianos, especially if you have some scripts or tools to help with the record-keeping. The only issue with big sample sets is the number of samples, which you have to deal with regardless of what tools you use. If you have to tweak filters for each sample, then it's a lot of work, regardless! But a well-recorded piano doesn't need that.

KVRist
37 posts since 5 Mar, 2021

Post Thu Jul 29, 2021 4:51 pm

Since the OP was sampling a digital piano, I think the first thing I would do is to figure out how many velocity layers the original hardware uses. Unless it's a very high-end piano, I suspect it only uses 4-5 layers, at most. Either check the specs for that model, or perhaps better yet, listen carefully to all the layers you recorded, and see if you can hear when it shifts to a new layer.

If you can successfully identify distinct layers in the original, just keep the loudest recording of each layer. That should save a bunch of work processing redundant layers, as well as make for a less huge file size.

I agree that doing any looping on these files would require a massive effort. I like small instruments, so I loop my samples a lot, but that's generally when I'm only using 1-10 samples for an instrument.
For the longest notes it probably wouldn't even be a big deal to just have them drop off abruptly - in most performances it's rare to play much longer notes, and after 12 seconds any note will be pretty quiet. After all, a mellotron will only have 8 second long notes max, and never any looping. And those are often choirs and strings, and other things that don't decay to silence like a piano!

Ultimately, though, if I were new to sampling, resampling a digital piano may not be the most fun way to start. The return of investment may be a bit low, since ultimately you'll, at best, end up with a copy of a digital piano. And don't forget the hundreds of nice free pianos available online in places like Piano Book.

KVRer

Topic Starter

23 posts since 29 Nov, 2020 from HellSinKey, Finland

Post Thu Jul 29, 2021 10:00 pm

JeffLearman wrote:
Wed Jul 28, 2021 11:37 am
96 KHz is overkill. Almost everyone uses either 44.1 or 48. Cut the file size in half and nobody will notice the difference. Consider using free r8brain to convert the rate; Audacity can do it too.
Yeah, I don't know what I was thinking. I'm going to convert the samples to 48 kHz. Thanks for the tool tip, I downloaded the r8brain.

JeffLearman wrote:
Wed Jul 28, 2021 11:37 am
15 velocity layers is more than I ever use, but whatever works for you. I find that real pianos vary a lot more than digital pianos with velocity. I assume you recorded a digital piano, not miking a Yamaha.


I might have been a bit unclear about the piano I sampled. It's sampled real grand piano (I assume)... it's from Nord Stage 3 --> Royal Grand 3D XL. I don't know what to call Nord Stage 3.

I think it’s good to have many different layers (with piano), so it doesn't sound so mind numbing, so quickly. And many layers make it sound more lively.

JeffLearman wrote:
Wed Jul 28, 2021 11:37 am
Practically speaking, you can ignore the tails at first, and come back and fix them later if they bug you.
Yeah, I might at first ignore the abrupt tails, but since I want to learn this sampling stuff, I'll try the looping technique too at some point.

And if I end up writing a track using this piano I can go back to library to record it with the real Nord Stage 3. (never released anything)

Thanks for all the advice!

KVRer

Topic Starter

23 posts since 29 Nov, 2020 from HellSinKey, Finland

Post Thu Jul 29, 2021 11:18 pm

MTorn wrote:
Thu Jul 29, 2021 4:51 pm
Since the OP was sampling a digital piano, I think the first thing I would do is to figure out how many velocity layers the original hardware uses. Unless it's a very high-end piano, I suspect it only uses 4-5 layers, at most. Either check the specs for that model, or perhaps better yet, listen carefully to all the layers you recorded, and see if you can hear when it shifts to a new layer.
The piano is from Nord Stage 3 --> Royal Grand 3D XL. Earlier I tried to find info about how many velocity layers there are, from nordkeyboards.com, but there seems to be no such info on their site. I would assume that it's thoroughly sampled and it has various parameters which are affecting to the sounds by velocity. I'll try to listen the samples.
Last edited by Cypsilonib on Fri Jul 30, 2021 12:20 am, edited 2 times in total.

User avatar
KVRAF
13036 posts since 8 Mar, 2005 from Utrecht, Holland

Post Thu Jul 29, 2021 11:29 pm

Make a midi file with just centre C going in velocity from 1 to 127 in 1sec intervals (renders just over 2 minutes of audio)
Then you might hear where one velocity layer starts and another ends. If there is a different amount of layers per zone of the keyboard, it's likely the middle has the most of them.
We are the KVR collective. Resistance is futile. You will be assimilated. Image
My MusicCalc is back online!!

KVRer

Topic Starter

23 posts since 29 Nov, 2020 from HellSinKey, Finland

Post Thu Jul 29, 2021 11:34 pm

Oh, and about the Kontakt... I can't afford it for a long time (or I don't know if I even want/need it ever) because there are other things I need to get too, like a new desktop computer, for example.
Last edited by Cypsilonib on Fri Jul 30, 2021 2:22 am, edited 1 time in total.

User avatar
KVRist
344 posts since 16 Sep, 2005

Post Fri Jul 30, 2021 12:21 am

https://github.com/psobot/SampleScanner
Not for piano but useful SampleScanner

SampleScanner is a command-line tool to turn MIDI instruments (usually hardware) into virtual (software) instruments automatically. It's similar to Redmatica's now-discontinued AutoSampler software (now part of Apple's MainStage), but open-source and cross-platform.

Return to “Samplers, Sampling & Sample Libraries”