My New Rhodes: Anyone Down for a Large Free Sampleset?

[egbert]

Hey Hollowsun, the test files should reach your mailbox in a few minutes now (twenty or so at most). And your absolutely right about the signal to noise ratio especially in regards to the decay of the note. It was hard to tell at some points where the note decay ended underneath the noise (which was probably flooring at around the -50db range).

Funkybot, here is a quote from the Scarbee site about the recording of Scarbee Rhodes:

The piano sound was captured directly from the harp, bypassing the passive tone control on the front panel, in order to capture the full frequency range of the instrument and minimise unwanted noise.

I'm not sure how easy this approach would be to emulate but it would certainly clean up the noise. If the pickups are single coil so would keeping the rhodes away from computer monitors and dimmers.

Here's the link to Scarbee's page on the Rhodes:

http://www.scarbee.com/products/rsp73/index.php

[hollowsun]

Ok.....

Funkybot has sent me the first few test samples (C2, C3 and C4 with two velocities) and I have done a quick job on them - half an hour of tweaking and programming on my S5000.

Unfortunately, the samples were a tad noisy but Rob (Lovesign of this parish) has run 'em through Adobe Audition to minimise this.

I've done a quick MP3.

Before you listen, bear a few things in mind:

* The playing is utter shite - it's just me doodling (and very badly at that!!!!)

* This is just *THREE* samples at C2, 3 and 4 mapped out accordingly

http://216.55.137.24/hollowsun/cd_mp3/crap_rhodes.mp3

Not for critical review or comment - just a 'taster'

Shows promise I think


Steve

[hollowsun]

Here's another done by Rob/Lovesign....

http://216.55.137.24/hollowsun/cd_mp3/FunkyBot.mp3

Again, nothing outstanding - the piano at the intro is a little dodgy coz it uses a run of the C4 sample well out of its range and the main comping is a little 'dead' but bear in mind again that this is just *THREE* samples and was knocked up without effects or production.


Steve

[griels]

Damn, that's rather funky Mr Lovesign... Didn't know there was a fellow jaather in town

[precisionsound]

Some tips to shrink it further. Loop the samples and save around 30% depending on which tone/key. Do a Sfz version with .ogg samples and the download will be aprox 10 times smaller.

Lars

[learjeff]

Anyone who thinks that a good Rhodes sample set can be done with one or two velocity layers is sorely mistaken -- obviously someone who isn't a serious player of the instrument.

My rhodes soundfont has 5 layers at the bottom end and that is woefully insufficient. I will probably do it again someday, but not in the forseeable future.

Having sampled my Rhodes and created what most people say is the best free Rhodes soundfont on the internet, here is my advice.

1) At the bass end, sample as many velocity layers as you can stand.

2) Have more layers at higher velocities, because the timbre changes more at high velocities than at low ones. (My samples were evenly spaced, 6dB apart.)

3) By Middle C, 4 or 5 layers are plenty -- 3 may even be enough. And you can lose a layer per octave going up from there. More than one layer in the top two octaves is unnecessary. Unfortunately, this doesn't save a whole lot of space because the higher notes are much shorter.

4) Record in 24 bits. Let sound designer reduce to 16 bits if so desired. I agree with Hollowsun that 16 bit samples are plenty. However, recording in 24 bits allows more headroom and "tailroom", so that if the recording isn't set up perfectly, the sound designer has more flexibility. (Note that the gain structure in the analog chain is still important.)

5) The Rhodes has a fairly low S/N ratio, but it's easily cleaned up in software, leaving lots of nice high frequency material in the samples. I find that any artifacts caused by this process are far less noticeable than the noise.

6) Sampling every 4th white key was ample. Variations that are noticeable at sample boundaries are more due to odd notes than note stretching and timbre changes. So, when choosing the notes, be sure to avoid the oddballs and stick to the good clean ones. Of course, feel free to sample every key or every white key. (Or every black key -- since they tend to be less used, you may be less likely to have oddball notes.) But every 4th key is ample, especially for a reduced-size soundfont.

7) Full length samples at the bottom end are over 30 seconds. I stopped mine after 25 to 30 seconds. Feel free to record the whole length, but rarely is that wave data ever used. This is a good place to compromise.

Looping works well if carefully done and if the decay parameters are carefully adjusted for each zone. The ideal result has no looping, but this is where you save the most space. My unlooped soundfont is 75M, the looped one is 12M.

Here's the kicker.

By far, the very hardest part of creating a good soundfont is recording the notes using consistent velocities for a velocity layer. By modifying the software I wrote to assemble my soundfont, I probably could make it work using more or less random velocities, using sample levels to automatically assign layers. This would be very tedious to try to do manually.

Someone mentioned that 24 bits compresses more than 16 bits, which I find is false when using lossless compression (and with lossy compression, why use 24 bits at all?)

Even using the best lossless compression (APE format), the resulting files will be pretty huge. However, if someone offers space for this project, I'll also be willing to assemble a soundfont.

My software does the trimming and detects the pitch to name each sample according to pitch (note name and MIDI note number). Currently the velocity layers are done manually -- I sample one layer at a time and so each wave file is one layer. I would have to expand the software to measure the peak RMS level and then assign velocities according to configured level/velocity curves (er, somehow).

[learjeff]

BTW, you can try my software yourself, it's free. But it's nerdware, meaning you have to read some text and run programs from DOS command shell -- it's not a GUI-based program. For more info just visit my site (below).

[Sascha Franck]

Autotrimming and autonaming you say, Jeff???
Wow, I need to have a look at that! Get someone do an interface, I'm such an idiot when it comes to any text based editing

Apart from that, I tend to agree with the velocity layering. If there's one trademark sound of a Rhodes, it's those "barking" notes at higher velocities. If you'd only sample those, no velocity to sample start modulation would get the thing right, and in case you'd leave them out, the best you'd get would be a rather mellow e-piano.

[Sascha Franck]

Oh, and please add some autolooping feature to your software

[egbert]

Hi Jeff - glad you dropped in on this thread.

4) Record in 24 bits. Let sound designer reduce to 16 bits if so desired. I agree with Hollowsun that 16 bit samples are plenty. However, recording in 24 bits allows more headroom and "tailroom", so that if the recording isn't set up perfectly, the sound designer has more flexibility. (Note that the gain structure in the analog chain is still important.)

5) The Rhodes has a fairly low S/N ratio, but it's easily cleaned up in software, leaving lots of nice high frequency material in the samples. I find that any artifacts caused by this process are far less noticeable than the noise.

Agreed - I think that if you are going to apply noise reduction software to a set of samples you would be better to let that software loose on the 24 bit samples and dither to 16 bit later if you need to save space.

Someone mentioned that 24 bits compresses more than 16 bits, which I find is false when using lossless compression (and with lossy compression, why use 24 bits at all?)

I would have to run a test on some piano samples to confirm or deny this (can't do it right now) but in the past I found that samples with an initial attack and a long decay down to very low levels compressed more than full volume song bounces for example. This makes sense since most of the tail is at much reduced volume and has fewer bits of information - the rest is zeroes. Analogous to compressing a bitmap that is mostly white background.

Thanks for sharing your software tools and thanks again for the SFs - really worth having.

Regards,
Eg

[Funkybot]

Just a quick update:

You know, this whole sampling thing is a pain in the ass . And I'm just talking about the recording, not even the programming. I would not have the patience to do this professionally.

Ok, so I think I got the first octave, these samples will definately need a good cleaning. Here's something I really like about my real Rhodes that physical models never bother to emulate: how undynamic some notes can be while others can be quite dynamic. For instance, on my Rhodes, B2, not a very dynamic note, the difference between a medium hit and a really hard hit is minimal, whereas just a few notes away at D2 is an extremely dynamic note. Sure I could take my Rhodes into manhattan and have it serviced by a place that specializes in them, getting that fixed, but then I'd be affraid I'd loose some of that "lived in" feel.

Mind you this is a very nice Rhodes, in great shape, but it's quite far from perfect, and while I'd like to sample it and make a nice freebie available, I don't want the most pristine Rhodes sound ever (between Mr. Ray 73 and Lounge Lizard, I already had that and didn't need to buy this one). If anything I'd want this to be a sampled Rhodes "warts and all," closer to what you might actually find, as opposed to one you'd buy in good shape then spend another $500 just servicing to make it perfect. I'm hoping that by the time this is done people will get that. In fact the only thing that I'm for as far as cheating to clean up the sound, is the removal of the unnessary noise (and boy is this thing noisy) at the decay of the notes, as otherwise it'd be quite bothersome once you started using up the polyphony. Oh and my F2 is actually an F3, I guess E2 or F#2 would have to serve as a replacement there (though it would be kinda funny to just leave it in, but I think Hollowsun would probably be against that one).

What I'm doing now is sampling the same note at about 4-6 velocities. I intended to do a few less, but after having done so it seems that you really need this at times, and I figure if anything Hollowsun could pick the best ones if he wanted to use less, but I'd rather give him the option of more. I'm leaving myself plenty of headroom, but on note decays the noise is quite prevelant, and I'm hoping some noise reduction software will help (the noise seems fairly consistent in it's sound, so it shouldn't be too tough to remove). I'm also recording 24 bit, but I imagine the final set will be 16, which actually makes sense, but I wanted to use 24 bits for the recording at the very least. Anyway I hope to get this done sometime mid next week (if not sooner) at which point I'll probably mail a CD of the files off to Hollowsun and let him take it from there.

Anyway, just wanted to chime in and let people know what was up. I really hope we can get a nice Rhodes set out of this. Personally I'm an EP freak myself and can never get enough different Rhodes/Wurly sounds.

[realmarco]

dunno if anyone said something like this but hows about you just recording each note at about 5 saples per note(every note) and this at each velocity (pp,mf,ff)X5 each

if its too much of a drag to programm why note just makes the raw wav files available to anyone

similarly to what iowa University did

just long files that we cut up on our own, so you can concentrate more on the just the quality recording rather then going through the whole process and then getting frustrated and ending up giving it all up

btw like someone said recordyour rhodes with''it'' as far as possible from the screen and fader lights...also experiment with placement of 'it' kinda like a single coil guitar(you move around and find the sweet spot where there is minimal noise)

also dont and I repeat dont do that thing as someone mentionned...you know setting the gain for each note differently...

keep it as natrual sounding as possible hammer hard on a loud note and set the preamps gain so that is just a fraction below distorting

oooh oooh also sustain your notes as long as possible I can always tell when looping is done

I just want the raw wav files I don't care about hardware samplers(which are useless to almost everyone hear at VSTi world)

[Funkybot]

Marco I was thinking about just trying to find somewhere to put up the raw wave files, but A) I'd have nowhere to host it, and between size and bandwith that would get pricy, then B) I'm not sure if anyone's even done a complete set out of those samples. And 5 takes at 5 velocites would just be a tad bit much, I think I'd go nuts and just off myself if I had to record all that .

[learjeff]

Right, Eg -- samples shrink down way more than music does, since they're far simpler harmonically. I particularly like APE format since it's lossless and silence shrinks to zero space. (True silence that is, not just "quiet". To get this benefit, you use a DAW or wave editor and select the silence between notes and silence it completely.)

Funkybot, you're right that the recording process is the hard part. It's particularly hard when you're trying to do it like I did, trying to get the same velocity for each note as I went up (then starting over at a new velocity).

Realmarco, what I find is while I can play ppp, pp, p, etc., up to FFF, while playing a piece, I'm completely unable to walk up and play a single note at a given velocity. It's really hard as hell. It reminds me, I had a friend who was a world-class percussionist (classical). She said that one of the coveted positions was playing Cymbal. I thought that was strange -- you only play one note. Her reply: "Exactly! and it has to be perfect." (Of course, it's more complicated than that, but you get the point -- it's HARD.)

Anyway, Funkman, you have the right idea, go for it. Start near the bottom and strike that key HARD. Make sure the record level goes high but doesn't peak. Then your levels will be set for the rest of the recording. (Test this by hitting a number of keys in the low register that hard and making sure none of them clip. On my Rhodes, the highest peaks were near the bottom, though I don't remember HOW near.)

Start the recorder going, and start at the bottom and play from louder to softer -- but don't worry about mixing them up in order. In other words, if you play lighter or harder than you intended, just let the note finish and try again. Also, feel free to abort any note by stopping within one second -- otherwise, let it ring.

For a Rhodes, especially at the low end, I feel it's FAR more important to sample lots of velocities than to sample lots of notes. On the other hand, if an idea works out, I might be able to get the software to select the best samples to cover the velocities and winnow out unnecessary samples -- the user would control how many total samples and the software would pick the best set.

In any case, I'd encourage you to sample lots of velocities at the low end, especially in the loud range, rather than sampling every key. This way it will be a lot easier to come up with a smaller soundfont that still captures the beast in all its glory.

I'd sure appreciate it if you would send me a copy of that CD, and Hollowsun and I can have a bake-off, a good-natured one of course. No doubt I have plenty to learn from him as he's done way more soundfonts than I have.

Hollowsun is likely to finish before I do. I'm neck deep at work right now, and don't have time to devote to the programming required.

Sasha -- I doubt I'll auto-loop. I'd have to seriously educate myself in forier math to have a shot at that -- hmm, there's an interesting challenge! But no, not in the forseeable future. EXSC makes looping so easy, though, that I can loop a big soundfont in a couple hours. I'd have to loop a LOT of soundfonts before it would cover the time I'd take just to learn the math! EXSC also now has a feature to automatically set the decay start time to the end of the loops (at my request).

[Sascha Franck]

Funkybot, as said, there's allways the Auditorium to host larger sample files.