Finally, after almost a year hiatus, the Djembe soundfont has been rolled out. To recap, I had given up on it after getting far enough for my daughter’s backing track – still not very happy with the results.
Then, a fine independent producer doing a web search for djembe samples stumbles across my humble blog. He contacts me and says that if I finish the soundfont, he may well be interested in purchasing it.
Well, a paying customer was incentive enough for me. Together (him providing critique, me providing manhours in front of the computer) we got it whipped into shape.
The main changes that I made were making the soundfont mono and reducing the number of samples to 10-12 per articulation. The brought the font down to a more reasonable size. If I ever branch out into some form of sampler, such as Kontakt, I’ll bring the instrument back up to full size with 20 stereo samples per articulation.
All in all, I’m happy with the way it turned out. You can check out the web page or order a copy of it here:
I got the next two syllables in on Tuesday night – Gun and Dun. I saved the soundfont and tried to load it into the Creative Soundfont manager. No go. I’ve had this problem before with soundfonts with very large single instruments. This one is around 50 meg right now.
This is annoying on several levels. First, I have loaded soundfonts much larger than this without any problem. However, the individual instruments are much smaller. This is a single instrument at 50 meg. Second, it’s been my experience that if the sound card won’t load the soundfont, neither will Synthfont. Finally, fixing this right is almost as much work as starting over from scratch.
Well, I have to get something done soon if I want to use this drum in my daughter’s backing track for Decatur, so here’s the temporary fix – I eliminate samples. First I’m going to take out everything below velocity 85, because I’ll never use that anyway.
If that’s not enough, then I’ll start pulling out samples where they are thickest until the darn thing will load.
The long term fix (and commercially viable one) is a bit more involved. I still want to maintain as many velocity levels as I can – the NS Congas are 645 meg with 40 velocity levels. Apparently the samplers that NS uses, namely Kontakt and HALion, can handle much larger file sizes.
So the first part of the fix is to re-mix the samples, this time with a gate to clearly delineate the end of the samples and silence the space between them. By using waveknife this time, I hope to make cutting up the individual samples much easier.
Then we’ll decide just how many velocity layers are really needed. I’m thinking more like 10-12 rather than the 18-25 I’m currently using. Remember, I’ve put in 6 of 14 notes and already run out of memory.
My next step, however, is to get this version working and make the backing track. I’ll post a sample when it’s ready.
A couple of quick things that I learned tonight – I can’t figure out how to delete samples from an instrument in Synthfont Viena – I accidentally imported all of the Go and Do samples twice. I e-mailed the creator to ask, but then found out that it is quite easy to do in Creative Vienna – just highlight the sample in the instrument and hit the delete key.
I also discovered an easy way to assign key range and root key in Synthfont Viena. In the instrument list right-click on “Global” and choose “Arrange Instrument Splits”. This brings up a dialog of all the samples in the instrument and a keyboard. You can left click and drag to select a range of keys for the assignment, and the right-click on a key to specify the root key. You should also be able to assign the velocity range from this screen, but I couldn’t get it to take – that is, I would assign the velocities, only to have them revert to 1-127 when I came back to the sample. This would be nice because you can do both the left and right samples at the same time, cutting the amount of work in half.
More tomorrow, hopefully. If I can get Gun and Dun done, I will have a completely playable instrument, and maybe can lay down a track on Bethany’s song.
It was a busy Easter, so I didn’t get much done in the way of music or soundfont creation. I did work a bit tonight, so wanted to report what I have learned…
I’m now at the point where I’m turning all of my individual wav files into a cohesive soundfont. There are two pieces of software that I’m aware of to do this – Vienna and Viena. Vienna (with two n’s) is by the Creative folks who make the Soundblaster cards that use soundfonts in the first place. Viena (with one n) is by the Synthfont people – the software that plays soundfonts. Both are free.
As for which is better, so far I pretty much like Viena better. It has a grid editing screen where you can make changes to the individual parameters of your samples fairly easily (although being able to change a mess of samples at once would be even better). The only advantage to using Creative Vienna that I’ve found so far is an easy graphical interface for setting key and velocity assignments. This makes it easy to see if you’ve screwed anything up. In Viena you have to do all this numerically. This will become important as you continue on my journey with me.
I started with the Go and Do syllables (see here for layout) and have just now realized as I am writing this that I screwed up the key assignments. My thinking was that I would start with the most commonly used syllables (which are really pa and ta) and get them working as practice. That explains why when I played my results they sounded funny. I was playing the tone thinking I was playing the slap. Makes sense now.
Anyway, what I learned tonight is applicable, regardless.
First, what I learned about assigning the keys – they must be a range, even if the range is only one key – i.e. 53-53. Velocities are similar – range even if it’s a range of one.
Once I got this figured out and tried to play my rudimentary soundfont, It just didn’t sound right. The left hand was playing really low and the right hand was playing really high. What was up with that?
Well, the second thing I learned tonight is that each sample has an assigned “start key”, which is what the sample player assumes is it’s native pitch. It then pitches up or down from there. I’ll have to see if you can set this pitch when you import the samples, but my program thought they were all middle C, which means the left hand was transposed down half and octave, and the right hand up by the same amount.
I got the start key straightened out, and tried playing the font again. Now the left hand was playing perfectly, but the right hand was still screwy – sounded wrong and occasionally I got double pitches. Creative Vienna to the rescue (I told you this would come in handy). Using the graphical interface I discovered that one of the samples was mapped to the entire velocity range (producing the odd overlaps and double triggering). I remapped the samples and opened back up in Viena, to discover that many of my “start key” values had disappeared. Once I got them back in, the font played exactly as I intended, except for that whole wrong syllable thing.
If I can get to it this weekend, I’m going to see if I can import and map another syllable. My fear is that I can only import samples into a patch once, and will have to start over with everything I did tonight.
At least I know how to do it now (which was my purpose in the first place). I’ll let you know…
Well, I’ve just performed a velocity test playing a standard MIDI set of congas. When I laid back, almost none of my velocities were under 90. When I played more aggressively, I peaked the meter at 127 numerous times. My digital piano has three velocity curves and these were the results for the most aggressive curve (have to hit it the hardest to peak the values).
I’m thinking this: I’ll listen to each of the sample sets and decide which is the first usable sample in terms of general loudness, and put that at 85. Then I’ll evenly divide up the remaining samples between the remaining values up to 127.
Last night I finally finished the “slicing and dicing” of the samples. I didn’t keep accurate track (I wish I had) of how long it exactly took, but it was on the order of 6 hours. This does not make it more likely that I will be doing the reverb-added version any time soon. I’ve put off the music far too much for this.
So, what is the next step? It’s time to assign the samples to the keyboard, and figure out how to set up the velocity switching.
In review, the djembe makes the distinct tones, Pa, the slap, Go, the tone, and Dun (Doon), the bass. Adding a variant for the off-hand we get Pa, Ta, Go, Do, Dun and Gun (Goon). We also recorded muted version of each of these (where the percussionist doesn’t remove his hand from the drum head, thereby not allowing the instrument to ring – although it does ring, just in a different way), and a variation on the slap where instead of lifting his hand straight up after striking the head, the percussionist continues the path of his hand down and behind the drum head.
Here’s what I’m thinking about for the layout of the keys (click to enlarge):
What I really need to figure out is how to assign the various samples in terms of key velocity – linking how hard the key is pressed to which sample plays. I’m thinking some sort of logarithmic scale weighted towards the top end. This is actually fairly self-serving because I play really hard and will otherwise never hear most of the samples.
My play is the play a but on the keyboard like I’m playing the djembe samples – maybe on a standard drum set congas – and see where the velocities lie. I’ll shoot for some soft playing and some more aggressive playing.
Thoughts are welcomed, as always.
I began slicing up the individual samples tonight. It’s going to take a while – two hours worth of work and I’m not quite 1/4 of the way done. I’m using a freeware program called Audacity to do the audio editing. It’s pretty boring work, but I observed something that was fascinating, and I thought I would report it.
So far I’ve completed the off-(left-)hand of bass and palm and their respective mutes. What I found was that the muted bass sustained longer than the non-muted bass. What?
Well, the sustain is a very small oscillation that continued until Ken lifted his hand from the drum head (which created an audible sound). The oscillation is not really audible, but it’s definitely there and regular in the waveform.
Lastly, I’m considering remixing the bass mutes with the main slider down about 10%. I’m ending up with quite a few samples clipping at the top end of each series. I’ll decide for sure after I listen to them – they may be audibly different. the clipping is short enough (<10ms) that it’s probably not audible, but since all the samples peak at 1.00 I can’t rank order them like all the others…
I probably won’t post again until the slicing and dicing is done, but until then I’m thinking about how I want to map these samples on the keyboard.
When the recording was complete, we had 14 different drum hits and one sound effect, each with 10-15 samples of increasing volume, each with 3-4 takes. That’s a lot of samples. The first step in mixing these samples down was to come up with a consistent mix and set of effects. We recorded a series of drum patterns to help with this.
So I sat down at my computer, set the sequence to loop the first eight-bar pattern, and began tweaking and playing. First I listened to each of the mikes seperately. That was very interesting. The overhead mike was very ambient with a lot of highs and definition, but no lows and little sustain. The side mike picked up a lot of hand-slap, what I can only describe as a “papery” sound – not sure if that’s helpfull. The bell mike, of course, was rich and full, but without definition.
All as expected, and all ready for mixing.
I asked for some advice in mixing from the folks at kvraudio forums. If you take a look, you will see that I got little help beyond “dry,” so that’s where I started. The best dry mix I came up with was the bass as 0 dB, the side at -5 dB and overhead at -10 dB. This maximized the tone and minimized the papery swooshy sound that I didn’t like. I also ran the main fader at about 100 (out of 127) to keep the peaks from clipping.
I played with several EQ and compression settings, and ended up with the Kjaerhus Classic Compression on the overhead and side mic. The overhead is set for a threshold of -16 dB and compression of 2.5:1. The side mic is set for a threshold of -8 dB and compression of 4:1. I tried adding minimal compression to the entire mix, but that killed the bass for some reason. Most likely I don’t know how to use the compressor.
I only left EQ in two places – first on the bass channel rolling off the highs a bit. Then I applied EQ to the entire mix, using subtle changes until I had dialed it in. To help, I started with a flat setting and essentially soloed each slider by taking it all the way up to see what happened.
As I was listening to the overhead soloed, I realized that it would be the perfect place to added a little space and ambiance. I have numerous reverb plug-ins to play with, but the two I use the most right now are the CDReverb that came with Power Tracks, and the Kjaerhus Classic Reverb, which is rapidly becoming my favorite. I used the supplied preset “016 Percussion”, unable to improve on it.
I polled the users at both PG Music Forums and kvraudio forums and in both places, the response was overwhelmingly for providing the dry version. I’ll do that and provide the wet version if the dry one doesn’t take forever to create.
Samples of the rhythms with and without the reverb can be listened to here:
Next, Slice and Dice.
The first part of creating a brand new soundont is the collection of the raw samples. In this case I wanted to sample an African drum called a djembe. I love the variety of tones that can be achieved from this drum in the hands of a skilled player. I’ve also had terrible luck finding one (a soundfont, that is) that I can use on my music system. Necessity being the mother of invention, I decided that it was time to create a high-quality djembe soundfont for my personal use and possibly for sale.
The first step was getting the personnel together. Two of my closest friends joined me in this activity. My friend Ken is a percussionist who owns a high quality 14″ Paulo Mattioli djembe (specifically, the one on the left). The other is my friend Will, an engineer with plenty of musical credits, himself. He also happens to own higher quality recording equipment than I do, plus I trust his ears and musical sensibility.
We met at Will’s house at 10:00 on Saturday morning and set about arranging the microphones. All three of us had headphones on and Will moved the mics about while Ken played a repeated note. When we settled on a position, we moved the mic stand up to it and locked it into place. Here’s the final set: Djembe Miked Up.
Here are the details: The overhead mic is a Behringer B-5 with the wind screen on. Without it we found that we were getting wind noise when Ken lifted his hand up off the drum head on the harder hits. It is set to a flat response and a reversed polarity.
In the bell of the drum was another B-2 Pro, this one set to omni, and direct to the mixer. Anything above line level was clipping the signal. After we finished recording, Will wondered what would have happened if we flipped the phase, but it was too late. I guess I’ll give that a try in mixdown and see what the results are.
All the signals were simultaneously recorded through a Behringer Eurorack mixer onto a PC-DAW running Power Tracks Pro Audio. Since I also use this software, we were able to burn a CD of the rough samples for later mixing in my own studio. All this took about 1 1/2 hours and we broke for lunch.
Now that we had the mics all set up, we had to decide what we were actually going to sample. Fundamentally, the djembe makes three distinct tones – the bass, the palm and the slap. Each of those has numerous variations depending on whether it is on- or off-hand, muted, or for the slap whether the hand rebounds up or slaps off the side of the drum and continues on down.
We also needed to make these samples at different loudnesses, so that when you hit the key harder to play the sample, not only does the volume increase, but the tone changes in the same way that the tone of the real drum changes when you play it harder. We set up a dB meter next to the drum with the intension of measuring and reproducing the volumes, but that proved unworkable.
What we ended up doing was Ken simply played a series of 11 – 15 hits at steadily increasing volume. We did this in 3 or 4 takes, so we ended up with 33-60 individual samples in one long wave file for each type of hit. Sometimes the hits smoothly increase, sometimes there’s a jump or even a small reversal, but with 3 takes we covered everything we should need, and will probably throw over half of it away as duplications.
After taking samples of all the hits we recorded Ken scraping his hand across the drum head in various ways. Finally, we recorded four drum patterns. These will have a few uses. First, I will use the patterns to mix the samples, and then apply the same mix and effects to all of the samples we took on Saturday. Second, we will try to duplicate these patterns in a sequencer with the mixed samples to see how good a job we did.
So, I have decided to host my production diary on-line. You have probably gotten here from a link on my home page. If not, welcome, come on in and sit a spell. The idea is that I’m going to keep a running log of my music production projects here on my WordPress blog. Since I never ended up using it for family updates as originally intended, this seems like a reasonable use for it.
Anyway, here’s the plan. I will post an update when I work in my studio, with a description of what I accomplished, what stood in the way of my accomplishing it, and how I worked around it. Maybe others who are trying their hand at home recording will read what I’m working on and be inspired. Or will drop me a line and inspire me. I also intend to post the occasional article about technical subjects that I learn about on the way, like VSTs or soundfonts.
Since I have ongoing projects right now, and have been keeping my production diary in a spreadsheet, the first entry for each song will include all the diary entries up to date.
Today I met with two of my closest friends and we recorded samples of a djembe all day. For those who don’t know, a djembe is an African drum that can play a variety of tones. I’ve never come across a decent sample of one I could play with my computer, so today we started the process of making one.
It took nearly two hours to set the mikes up and get them dialed in. Once that was finished and we had full tummies (lunch), the actual recording went fairly quickly. We took 2-4 takes each of 6-15 samples of increasing volume of each of the tones the drum can make. The intention is to build a soundfont from these samples. More on soundfonts later.